repositorio yachay tech: home - universidad de ......año en yachay tech, quien me permitió ser su...
TRANSCRIPT
UNIVERSIDAD DE INVESTIGACIÓN DE
TECNOLOGÍA EXPERIMENTAL YACHAY
Escuela de Ciencias Matemáticas y Computacionales
TÍTULO: Non-predictive and predictive models to recognize
the learning style of the students: A case study
Trabajo de integración curricular presentado como requisito para
la obtención
del título de Ingeniero en Tecnologías de la Información
Autor:
Torres Molina Richard
Tutores:
Ph.D Guachi Guachi Lorena
Ph.D Ortega Zamorano Francisco
Urcuquí, agosto 2019
Dedicatoria
Dedico este trabajo de integración curricular a DIOS y a mi madre, ALMIRA MOLINA, por
creer siempre en mí y estar presente en los momentos más difíciles de mi vida. A los profesores
por su apoyo en mi formación como ingeniero, y a mis compañeros de clase en esta etapa de
mi vida.
Richard Andrés Torres Molina
Agradecimiento
Quiero agradecer a mis supervisores de investigación Lorena de los Ángeles Guachi Guachi
Ph.D., y Francisco Ortega Zamorano Ph.D. en el desarrollo de la tesis. Asimismo, al director
Juan Vázquez, a la maestra Silvia Díaz y al personal administrativo de la escuela “Teodoro
Gómez de la Torre” (Ibarra-Ecuador) que han contribuido en los datos recopilados de este
trabajo.
Al mismo tiempo, me gustaría agradecer a mi comité, incluidos los profesores Zenaida del
Castillo Ph.D., e Israel Pineda Ph.D. La profesora Zenaida Castillo fue mi profesora de primer
año en Yachay Tech, quien me permitió ser su asistente de investigación en la implementación
de cuestionarios en Maple T.A. a los estudiantes de nivelación. Aunque no he tenido la
oportunidad de trabajar con el profesor Israel Pineda, sus clases me han ayudado a estar
preparado en mi carrera para elegir el lenguaje de programación adecuado en un contexto de
investigación o empresa. En estos cinco años, tengo que agradecer a muchas personas, a el
profesor Rigoberto Fonseca Ph.D., el profesor Ernesto Ponsot Ph.D., Gerardo Alvarado Ph.D.
y al personal administrativo. A la vez agradezco a mis amigos y compañeros de clase,
especialmente a Andrés Riofrío y Andrés Banda. Yachay Tech, fue una de las mejores
decisiones que he tomado en mi vida, y estaré siempre agradecido por la oportunidad de ser
parte de esta comunidad.
Richard Andrés Torres Molina
Resumen
Por la falta de personalización en la educación, los estudiantes obtienen un bajo rendimiento
en diferentes materias en la escuela, particularmente en matemáticas. Por lo tanto, la
identificación del estilo de aprendizaje es una herramienta crucial para mejorar el rendimiento
académico. Aunque los métodos tradicionales, como los cuestionarios, se han utilizado
ampliamente para la detección de estilos de aprendizaje en jóvenes y adultos por su alta
precisión, produce aburrimiento en los niños y no ajusta el aprendizaje automáticamente a las
características y preferencias de los estudiantes con el tiempo. En esta investigación, se han
utilizado cuatro técnicas, dos modelos no predictivos de estilos de aprendizaje: cuestionario
CHAEA-Junior y Mini-Juegos; y dos modelos predictivos de estilos de aprendizaje: Redes
Neuronales Artificiales y Redes Bayesianas. En primer lugar, para identificar los porcentajes
en los estilos de aprendizaje en cada estudiante, se utilizó el cuestionario CHAEA-Junior y las
preguntas matemáticas de los Mini-Juegos (Competidor, Soñador, Lógico, Estratega) basados
en la teoría de aprendizaje de Kolb. Luego, los datos recopilados del cuestionario y los Mini-
Juegos se usaron en los dos modelos predictivos. Las pruebas experimentales muestran que la
herramienta óptima en el reconocimiento general del estilo de aprendizaje para los estudiantes
de la escuela “Teodoro Gómez de la Torre” (Imbabura - Ecuador) son los Mini-Juegos basados
en estilos de juego ADOPTA, seguidos de Redes Neuronales Artificiales y Redes Bayesianas
para el reconocimiento del estilo de aprendizaje, como instrumentos de investigación para
brindar un aprendizaje personalizado a los estudiantes ecuatorianos.
Palabras Clave:
Detección de estilos de aprendizaje, Reconocimiento Automático, Red Neuronal Artificial,
Video Juegos.
Abstract
By the lack of personalization in education, students obtain low performance in different
subjects in school, particularly in mathematics. Therefore, learning style identification is a
crucial tool to improve academic performance. Although traditional methods such as
questionnaires have been extensively used to the learning styles detection in youths and adults
by its high precision, it produces boredom in children, and it does not adjust learning
automatically to student characteristics and preferences over time. In this research, four
techniques have been used, two non-predictive learning styles detecting models: CHAEA-
Junior questionnaire and Mini-Games; and two predictive learning styles detecting models:
Artificial Neural Networks and Bayesian Networks. Firstly, to identify the percentages in
learning styles in each student, it was used CHAEA-Junior questionnaire, and mathematical
questions from the Mini-Games (Competitor, Dreamer, Logician, Strategist) based on Kolb's
learning theory. Then, the gathering data from the questionnaire and the Mini-Games were
used in the two predictive models. The experimental tests show that the optimal tool in the
overall learning style recognition for students of the school “Teodoro Gómez de la Torre”
(Imbabura - Ecuador) are, the Mini-Games based on ADOPTA playing styles, followed by
Artificial Neural Networks and Bayesian Networks for learning style recognition, as research
instruments to provide personalized learning to Ecuadorian students.
Key Words:
Learning Style Detection, Automatic Recognition, Artificial Neural Network, Video Games.
Contents
1 Introduction 5
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Scope of the thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Dissertation overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2 Theoretical Framework 10
2.1 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.1 Non-Predictive Learning Styles Detecting Models . . . . . . . . . . . . . . . . . . . 10
2.1.2 Predictive Learning Styles Detecting Models . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Learning Styles Detecting Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Non-Predictive Learning Styles Detecting Models . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3.1 Questionnaires . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.2 Video Games . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4 Predictive Learning Styles Detecting Models . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4.1 Artificial Neural Networks (ANN) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4.2 Bayesian Networks (BN) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3 Techniques for Learning Style Detection 32
3.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2.1 CHAEA-Junior Questionnaire (CHAEA-JQ) . . . . . . . . . . . . . . . . . . . . . . 34
3.2.2 Mathematical Mini-Games . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2.3 Artificial Neural Networks (ANN) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2.4 Bayesian Networks (BN) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3 Experiment Set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.3.1 Instruments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.3.2 Participants and procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.3.3 Data Preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.3.4 Quality Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.3.5 Experiment Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
1
4 Results, Discussion and Conclusion 54
4.1 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.2 Restrictions for high precision in learning style identification . . . . . . . . . . . . . . . . . . 59
4.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
References 63
Appendices 67
A Appendix 1. 67
2
List of Figures
1 Kolb’s Learning Styles [1]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2 Relationship between learning styles based on LSI [2]. . . . . . . . . . . . . . . . . . . . . . 15
3 The four ADOPTA playing styles together with the learning styles of Honey and Mumford and
of Kolb [3]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4 General block steps to detect the learning styles. . . . . . . . . . . . . . . . . . . . . . . . . . 33
5 Detail steps to detect the learning styles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
6 CHAEA-Junior Block Diagram. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
7 (a) Architecture and (b) Deployment Diagrams in the Web Application. . . . . . . . . . . . . 36
8 Screenshot from the CHAEA-Junior Questionnaire. . . . . . . . . . . . . . . . . . . . . . . . 36
9 Screenshot from the Competitive Mini-Game Style. . . . . . . . . . . . . . . . . . . . . . . . 38
10 Screenshot from the Dreamer Mini-Game Style. . . . . . . . . . . . . . . . . . . . . . . . . . 39
11 Screenshot from the Logician Mini-Game Style. . . . . . . . . . . . . . . . . . . . . . . . . . 40
12 Screenshot from the Strategist Mini-Game Style. . . . . . . . . . . . . . . . . . . . . . . . . 41
13 Top-level architecture of the artificial neural network approach. . . . . . . . . . . . . . . . . . 42
14 Initial state Bayesian Network diagram. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
15 Bayesian Network updated after the student answers the question. . . . . . . . . . . . . . . . 45
16 Reflector Learning Style Percentage vs. Learning Styles Detecting Models. . . . . . . . . . . 56
17 Overall Error Percentage vs. Learning Style Detecting Models. . . . . . . . . . . . . . . . . . 58
3
List of Tables
1 Summary Table about questionnaire’s to learning styles detection. . . . . . . . . . . . . . . . 18
2 Summary Table about video games to learning styles detection. . . . . . . . . . . . . . . . . . 25
3 Summary Table about Artificial Neural Networks and Bayesian Networks to learning styles
detection. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4 Average of percentages learning styles based on CHAEA-JQ, Mini-Games, Artificial Neural
Networks, and Bayesian Networks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4
1 Introduction
1.1 Motivation
Every learner acquires information in different ways known as “learning styles” based on its cognitive, affec-
tive, and psychological factors. These factors determine how a person perceives, interacts, and responds to
the learning environment [4]. Being some recommendable activities depending on the learning styles: Activist
(brainstorming, problem-solving), Reflector (interviews, self-analysis questionnaires), Theorist (model analy-
sis, background information), and Pragmatist (case studies, problem-solving discussion). However, the current
learning system implemented in primary level schools, has a lack of providing customized learning to students
in different areas of knowledge, due to only a few educators have started to identify learning styles to improve
learning and teaching techniques. Notably, one of the challenges is given on the study of mathematics in stu-
dents worldwide [5], where different studies suggest that learning mathematics is difficult; therefore, students
performance is unsatisfactory [6]. Being Latin America with one of the lowest performance worldwide [7].
To overcome this limitation, static and automatic approaches for identifying learning styles have emerged over
time.
In the past years, questionnaires have been the most common approach to identify learning styles, char-
acterized by its excellent reliability and validity on undergraduate students, business administration students
and even doctors [8], [9], [10]. In any case, they have also been subjected to some criticism considering that
a questionnaire is a static approach, where their results are no longer valid over time, while learning styles
change continuously. As well as, in the majority of cases, the filling out a questionnaire produces boredom in
children. Besides, students are not aware of the importance of the survey for the future uses, which may tend to
pick answers self-assertively. Even in some cases, students can be influenced by the questionnaire formulation
to give answers perceived as more appropriate. To overcome its difficulties such as boredom, recent proves had
established a correlation between playing styles that match with learning styles applied on entertainment games
in education.
The learning style identification has also been investigated in technical fields like mechanical engineering,
for example in [11], [12] the impact of negative knowledge is discussed and implemented as a way to pre-
vent and improve competency in computer-aided-design modeling. In [13], practical experience is linked to
theory to help the novice in improving their theoretical results, and in [14] categories of skills and knowledge
are defined for defining questions and related significant scores. Besides, in the most recent years, some au-
tomatic approaches have been introduced in the learning style identification [15], [10], [16]: Decision Trees,
Genetic Algorithms, Artificial Neural Networks, Bayesian Networks, etc. Since automatic approaches tend to
be more accurate and less error-prone, they are focused on educational systems that adjust learning to student
5
characteristics and preferences over time.
Although numerous static and dynamic approaches for learning style identification have been introduced
with high accuracy, several primary educational systems in Ecuador, mainly for learning mathematics, is still
a challenge. Being the primary goal of this proposal to diminish the lack of adaptive learning in mathematics
primary educational level of Ecuadorian students through the identification of the most accurate technique to
detect their learning styles.
6
1.2 Scope of the thesis
This thesis aims to introduce the following contributions: a) Discussion about existing non-predictive learning
styles detecting models: CHAEA-Junior questionnaire and Mini-Games; and predictive learning styles detect-
ing models: Artificial Neural Networks and Bayesian Networks. b) Propose two new Artificial Intelligence
design models for automatic learning style recognition. c) Introduce two tools in non-predictive learning style
identification (web system for the CHAEA-Junior, and mathematical Mini-Games based on ADOPTA playing
styles). d) Comparison of the proposed four approaches: CHAEA-Junior questionnaire, Mini-Games, Artificial
Neural Networks, and Bayesian Networks.
The experiment took place with a group of children of 11 and 12 years old of seventh grade primary educa-
tion from the school “Teodoro Gomez de la Torre” (Ibarra-Ecuador). The gathering data from the questionnaire
and the Mini-Games were used in two Artificial Intelligence techniques proposed in this work, which are the
Back Propagation algorithm, and the Bayesian Networks. The data correspond to 100 students from the school
state before, divided randomly into two data sets, 80% for training and 20% for the testing phase. The four
methods, CHAEA-Junior questionnaire, Mini-Games, Artificial Neural Networks, and Bayesian Networks,
were evaluated on the testing set.
The CHAEA-Junior was implemented in a web system composed of 24 randomly questions from the 44
questions corresponding to the four learning styles: Active, Reflector, Theorist, and Pragmatist. The Mini-
Games were developed based on ADOPTA styles: Competitor, Dreamer, Logician, and Strategist. The goal of
the Mini-Games was to solve basic mathematical operations (sum, subtraction, multiplication, and division) ac-
cording to the rules in each Mini-Game. For instance: shooting rockets to the correct answer in the Competitor
style, pressing a puzzle piece to the right answer after hearing an avatar in the Dreamer Style, pressing a card to
the correct answer in the Logician Style, and with an avatar that jumps and sends bubbles to the correct answer
in the Strategist style. Each playing style has three levels: easy, medium, and difficult.
The percentages of learning styles are the “output” in the four techniques obtained from metrics in the
questionnaire and/or the Mini-Games. Meanwhile a particular “input”, is used in each model as follows: a)
binary answers captured from web site are used in the CHAEA-Junior questionnaire, b) the gaming scoring
achieved is used in the Mini-games at the first playing session, c) the CHAEA-Junior answers in Artificial
Neural Networks, and d) Mini-Game Score and Answer in the Bayesian Networks algorithm at the second
playing session.
The students took an initial mathematics test of 96 questions using the different Mini-Games stated before
with the various levels; then they answered 24 questions related to the questionnaire. The data was applied
in the Back Propagation algorithm to predict the student learning style. Finally, they played the Mini-Games
7
embedded with the Bayesian Networks algorithm in a 21 minutes game session.
The restrictions in this research were the sample size, room equipment, and time, reasons by it were possible
to extract data from only 100 students to be part of the experiment. So, this work is an observational study with
non-probabilistic sampling.
8
1.3 Dissertation overview
The thesis is organized in 4 chapters as follows: Chapter 2 presents the theoretical framework explaining
important concepts and studies related to learning style recognition with the use of static methods such as
questionnaires and video games, and dynamic methods such as Artificial Neural Networks and Bayesian Net-
works. Then, Chapter 3 depicts the description and analysis of the proposal chosen techniques, metrics used for
comparison purposes and experimental setup. Finally, Chapter 4 finishes with the obtained results, discussion,
restrictions for high precision in learning style identification, and conclusions remarks.
9
2 Theoretical Framework
A problem facing students in the classroom nowadays, is the lack of personalized learning in different subjects,
being one of the most challenging, the study of mathematics in students worldwide. As an outcome, the student
performance is deficient in that area, being Latin America with the lowest performance in comparison with
developed countries.
One of the keys to filling the gaps in the learning experience is the detection of learning styles in the student.
Therefore, specific models for learning style identification are used for decades. These learning styles detecting
models can be classified into non-predictive and predictive ones. The first ones related to questionnaires and
video games. The other ones related to prediction techniques based on Artificial Intelligence models to detect
the individuals learning styles. Thus, this chapter will be divided into two sections: critical concepts linked to
this research and learning styles identification models.
2.1 Concepts
2.1.1 Non-Predictive Learning Styles Detecting Models
• ADOPTA (ADaptive technOlogy-enhanced Platform for eduTAinment)
Family of playing styles correlated with learning styles based on Kolb’s learning theory. They are related
as: Competitor⇔ Activist, Dreamer⇔ Reflector, Logician⇔ Theorist, and Strategist⇔ Pragmatist.
• “Cuestionario Honey-Alonso de Estilos de Aprendizaje” (CHAEA)
Modification of the Learning Style Inventory, used in universities and translated to Spanish. This ques-
tionnaire classifies students as the Learning Styles Questionnaire.
• CHAEA-Junior Questionnaire (CHAEA-JQ)
Adapted CHAEA version to elementary and secondary education students, with ages between nine to
fourteen years old.
• Index of Learning Style (ILS)
Questionnaire to diagnose learning styles build on Felder and Silverman’s model. Where two learning
styles were defined for each of the four dimensions: Sensing /Intuitive, Visual/Verbal, Active/Reflective,
Sequential/Global.
• Learning Styles
10
Defined as the combination of many biological and experimentally characteristics that allow an individual
to perceive information and acquire knowledge.
• Learning Style Inventory (LSI)
Questionnaire created by Kolb to measure and understand a student unique individual style of learning.
The LSI identifies four distinct learning styles: Diverging, Assimilating, Converging, and Accommodat-
ing.
• Learning Styles Questionnaire (LSQ)
Questionnaire developed by Honey and Mumford, an instrument derived directly from Experimental
Learning Theory. It presents four learning styles: Activist, Reflector, Theorist, and Pragmatist.
• Questionnaire
A research instrument composed of a set of questions to find specific characteristics in a person.
• Video games
An electronic game with a user interface to generate visual feedback in a computer monitor, arcade
machine, or mobile phone.
2.1.2 Predictive Learning Styles Detecting Models
• Artificial Intelligence (AI)
Field related to how machines can learn and solve tasks in a trusted manner. It is divided into two groups:
Symbolic-Deductive Intelligence, and Computational Intelligence.
• Artificial Neural Networks (ANN)
Computational model in AI to tries to simulate the neural connections in the brain, to solve problems in
regression and classification.
• Back Propagation (BP)
An algorithm in AI that propagates back the derivative of the error function, from the end to the start
of the network. The weights are initialized randomly, where the error function gradient is applied to the
initial weights correction.
• Bayesian Networks (BN)
11
AI computational model with an acyclic graph that consists of nodes and arcs representing correlations
between variables, and the calculation of probabilities among them.
• Decision Trees
An algorithm used in AI composed by nodes and with arcs between them. It is used to classification
problems.
• Genetic Algorithms
A search-based optimization technique that is based on the process of natural selection in evolutionary
algorithms.
2.2 Learning Styles Detecting Models
Learning styles allow characterizing a student cognitively and psychologically. Different theories have detected
learning styles according to certain features. Hence, the following sections explore non-predictive and predic-
tive learning styles detecting models. Non-predictive models are integrated by questionnaires and video games,
whereas the predictive models related to AI models. So, it is provided a review of relevant works related to
each model.
2.3 Non-Predictive Learning Styles Detecting Models
A broad number of instruments have been developed to learning styles detection. In most of the cases are
static methods, being the most relevant, questionnaires and video games. Questionnaires are composed of a
set of questions based on cognitive styles theories [17]. The common questionnaires are Learning Style Inven-
tory, Learning Styles Questionnaire, Learning Styles Questionnaire, “Cuestionario Honey-Alonso de Estilos
de Aprendizaje” (CHAEA), and CHAEA-Junior Questionnaire. The answers given in these instruments allow
calculating the learning styles by the scoring in each one.
Although, questionnaires are the traditional method of learning styles recognition. It has some weaknesses
related to its static nature by time-consuming and boredom in students [18]. So, an innovate approach has
been developed by using video games [2]. Where the metrics in the game are retrieved from a web or mobile
system to recognize the learning style of the students respectively. For these reasons, it has been selected
the CHAEA-Junior Questionnaire, a questionnaire to testing purposes in children, and Mini-Games based on
ADOPTA playing styles.
12
Figure 1: Kolb’s Learning Styles [1].
2.3.1 Questionnaires
Curry [19] defines learning styles in how learners interact with, perceive and respond to the learning environ-
ment with specific cognitive, affective, and psycho-social behaviors. Since decades, various theories had to
emerge to identify learning styles, each one with a questionnaire. The first researcher to propose a way to
identify the learning styles was Kolb. He built the Learning Style Inventory (LSI), a questionnaire based on
experiential learning theory (ELT) created as a curriculum development project from MIT. ELT defines learn-
ing, as the acquisition of knowledge done by the experience, and it is related to how the individual assimilates
information and takes decisions. The ELT is perceived as a four-stage cycle as seen in Fig. 1, that is divided
on:
• Concrete experience (CE) or “feeling”: learning is determined by experience, where the individual keeps
excellent social relations.
• Reflective observation (RO) or “observing”: learning is determined by observation and listening.
13
• Abstract Conceptualization (AC) or “thinking”: learning in a theoretical and systematic way with the
adoption of concepts and theories.
• Active Experimentation (AE) or “doing”: learning by practice in doing work that requires decision taken.
The initial LSI [19] was composed of nine questions and extended to twelve questions in 1985. Each
question had four different options to be ranked by the respondent from 1 to 4. After the poll is answered, each
learning stage (CE, RO, AC, and AE), it is given a score. Then, by mixing the scores from the LSI as seen in
Fig. 1, it was classified the students in four learning styles as follows:
• Diverging (CE/RO): Individual that perceives situations with different points of view by listening, observ-
ing, and exchange new ways to solve problems. At the same time, they are eager to recollect information
with an imagination tendency and understanding to their peers. They are emotional with a tendency to
arts and cultural interest.
• Assimilating (AC/RO): Individual that prefers to use information logically and concisely. This kind of
people is more interested in ideas and abstract concepts rather than establish social relationships. They
are more theoretical rather than practical with preference to lectures, theoretical and analytic models, and
scientific careers.
• Converging (AC/AE): Individual that finds a practical approach in ideas and theories. These learners
have solving problems capacity and decision making. At the same time, they prefer to work on technical
issues (careers related to technology) rather than interpersonal relationships. The people with this style
wants to obtain adequate information in experimenting with new ideas, laboratory tasks, and practical
applications.
• Accommodating (CE/AE): Individual that learns with practical experience to new challenges. The people
with this style makes decisions based on “intuition” rather than logical analysis. They depend on other
people to obtain information that its technical analysis. They are more adequate to careers on sales and
marketing, with preference in teamwork to do tasks, work on the field, and test different methods to solve
a problem.
The LSI questionnaire has different versions since its creation from 1971 until 2005. LSI has been used in
art college students, psychology, and business students with strong reliability in diverse populations. The LSI
has been tested in teens and adults with not intended for younger children [1]. The LSI has been translated into
many languages including Arabic, Chinese, French, Japanese, Italian, Portuguese, Spanish, Swedish, and Thai.
14
Figure 2: Relationship between learning styles based on LSI [2].
Kolb’s work was examined by Honey and Mumford’s to create the Learning Styles Questionnaire (LSQ)
[19], an instrument derived directly from ELT. LSQ recognizes four learning styles based on people main
characteristics. Fig. 2 shows the learning styles, and are described as follows:
• Activist: Open minded individual to new experiences and challenges. The main activities for these
learners are brainstorming, and problem-solving.
• Reflector: Individual that consider the problem in different perspectives by analysis to obtain conclu-
sions. They tend to reflect the situation and listen before taking a decision. The main activities for these
individuals are interviews and self-analysis questionnaires.
• Theorist: This learner uses logic to build relations and incorporate all details into a problem with a
tendency to be perfectionists. The main activities for these learners are model analysis and background
15
information.
• Pragmatist: Learner that applies theories and techniques in a practical manner, whose ultimate goal is
to try on a real scenario. The main activities for these individuals are case studies and problem-solving
discussion.
The questionnaire was initially designed with 80 dichotomous items in four groups of 20 items for each
of the four learning styles. The respondent answered each question with “agree” or “disagree” with the state-
ment related to the learning style. This instrument showed excellent test reliability in learning style detection;
intended to academy and industry.
Alonso, Gallego, and Honey developed a Spanish version of LSQ, named “Cuestionario Honey-Alonso
de Estilos de Aprendizaje” (CHAEA). The CHAEA was designed to be applied in university students. This
questionnaire is consisted of eighty questions as LSQ with four groups of twenty questions, corresponding to
the four learning styles distributed randomly. To each one of the learning styles, there are a set of charac-
teristics associated skills. For instance, active people have a tendency with innovation, creativity, experience,
leadership, desire to learn, and idea generation. Reflective people are more receptive, patient, eager to research,
and observation. Theoretical people are logical with clear objectives to make models, concepts, and theories
connections. Pragmatic people are realistic, efficient in decision making, enjoy experiment simulations on real
problems.
Then, Felder and Silverman [19] interested in the performance of engineering students decided to develop
a learning model based on two successive phases: reception and processing of information. In the reception
phase, the senses captured by external information and internal information in a thoughtful way is accessible
in the individual by selecting the specific knowledge and avoid the rest. On the other hand, processing may
involve reasoning or memorization. In the Felder and Silverman model, two learning styles were defined for
each of the four dimensions (sensing / intuitive, visual/verbal, active/reflective, sequential/global). A more
broad definition of each dimension is defined as:
• Perception: Sensing individuals are more practical in solving real-world problems. They are careful
in detail and prefer to learn from evidence and explicit material. While the intuitive individual adopts
concepts and theories seeing connections creatively and innovatively.
• Input: Visual individuals learn by seeing, images, flowcharts, or diagrams. Verbal individuals learn in a
textual representation in an oral or written form.
• Processing: Active individuals like to develop experiments or implement them, with preference to talk to
other members of the group. Reflective individuals think individually in the learning material.
16
• Understanding: Sequential individuals that learn linearly. Global individuals that learn randomly and
even skip steps to find the solution.
Furthermore, it was created a questionnaire based on the Felder and Silverman [19] model named Index
of Learning Style (ILS). It is composed of 44 questions, where every question is focused on determining the
different learning styles of each. The scale for each learning style starts from the integer value between -11
and +11. For example, if a learner has a visual preference, it adds to the dimension related to the interaction of
information, while an answer for verbal subtracts to that dimension.
The majority of questionnaires are from testing purposes in adults, but to testing the learning styles in
children, it was created a new poll called CHAEA-Junior [20] to students in elementary education level and first
year of high school. This test detects the learning style of the students taken into consideration the psychological
children characteristics between nine and fourteen years old. The language was adapted to be understood by
the children and then tested to 258 students from Spain. At the first trial, it was selected 40 items from the
80 items from the CHAEA, 10 for each learning style scale. Then, for the second trial, it was increased the
questionnaire to 44 questions in a sample of 1594 students. The results proved the reliability of the CHAEA-
Junior questionnaire (CHAEA-JQ), and in a reciprocal way to the CHAEA itself. A summary related to the
literature review about the detection of learning styles using questionnaires is in the following Table 1.
17
Methods Main Application Pros Cons
Kolb Learn-
ing Style
Inventory -
Version 3.1,
[1], 2005
Identification for
learning styles
based on Kolb
Learning Style
Inventory - Version
3.1
Strong reliability in art
college students, psychol-
ogy and business students
Not intended for children
Questionnaires
based on
psycholog-
ical traits,
[19], 2013
Comparing differ-
ent questionnaires
to identify learning
styles
No available No available
CHAEA
Junior Ques-
tionnaire,
[20], 2014
Learning style
detection using the
CHAEA-Junior
Questionnaire
in primary level
education and
first levels of high
school
Test evaluated in children No available
Table 1: Summary Table about questionnaire’s to learning styles detection.
2.3.2 Video Games
Video games have been used as entertainment to young generations in most of the cases. Although, video
games [21] can be used in the educational field to improve cognitive abilities, visual short-memory, spatial cog-
nition, boot memory capacity, promoting arithmetic performance, and mathematical skills enhancement. These
benefits are primordial to primary level students, between the age of four and twelve years old. Hence, in Bel-
gium, it was taken into consideration how second graders react by using a mathematics test, working memory,
and visuomotor skills in pre and post sessions; a comparison between two methods with a mathematical video
18
game and paper exercises were taken into consideration. The experiment was done with 52 children divided
into two groups to test each method. During three weeks, the first group used the mathematical video game
named Monkey Tales, and the other group, math exercises in the paper. The module of the video used was
“Museum of Anything” determined to children in the third grade to test what the student learn in the second
grade. The game did not teach arithmetic skills, instead, motivated the student. It used an algorithm to be aware
of the learning curve and increase the difficulty of the exercises as it progresses. The mechanics of the games
were neutralizing a laser or shoot rockets. The presentation of the arithmetic problems and type of questions in
the mini-games were with the same level of difficulty than the paper’s exercises. At the same time, the students
were asked about how was the experience in the game with a scale of 12 items with components such as “great”,
“tiring”, “boring”, and “difficult”.
To the data analysis, it was calculated socio-demographic variables of each group, the variables group (gam-
ing vs. paper exercises), and session (pre-test vs. post-test). As results, after the mathematics test was done,
it showed that the students using video games obtained, working memory scores twice as better than the tradi-
tional approach. If the improvement was more significant in their working memory, the children finished faster
the post-test session. At the same time, there is a slight correlation between reaction times on the mathematics
test and working memory scores. The most attractive characteristic of this work is the students’ enjoyment
in using the mathematical video game in comparison with the children who practice traditional methods. In
spite of these findings, the researchers, due to the small sample size, failed in the correlation between enjoy-
ment scores and cognitive measures scores. Therefore, future research must be continued done to find evidence
between improvement in learning and enjoyment.
To enhance the mathematical skills in children, a group of researchers [22] proposed a serious Arabic game
to grant learning improvement in mathematics, and secure the pupils’ amusement. The video game main idea is
by single elimination of correct answers, a game environment, main character, and items in the game scene. In
the educational game, there is an adaptation by gaming rules to provide mathematical questions to the learner
with enjoyable activities in the gaming environment. For that purpose, three games were developed to promote
knowledge and competence. In this sense, the young pupils played the game with diverse content suggested
by the Ministry of Education from Saudi Arabia. The three games are: a) About the studied of integer and
decimal numbers (integer identification, multiple of 5 and 3, divisor of 9 and 4), by moving a helicopter to the
circles which serve as the right answers; b) About to perimeters and area calculations in geometric shapes; and
c) About units and measurement of lengths, weights, surfaces and volumes (multiples and fractions of each
measure and physical signification), where the right answers are selected by shooting them with a helicopter in
the game. The answers to the questions were represented with numbers on bubbles showing random trajectories
moved by arrow keys and mouse orientation. The game incorporated a starting number of lives that regulate
19
the numbers of intents with sound effects responses to boost the player. The player needs to follow a certain
amount of rules to accomplish the game goal. The Arabic game learning environment has shown that it is a
way to develop cognitive level and mathematical skills in children to increase learning growth. However, the
game itself did not pretend to be an automatic diagnostic system, but it was a diagnostic assistance system.
So, the majority of games has been applied to enhance the mathematical skills of the alumni; however, at
that moment there was no exist a learning style recognition technique to provide a custom learning experience.
Therefore, a group of scientists [23] investigated about the learning style differences and attitudes between dig-
ital game-based learning among users with mobile phones. To collect the data, they used different instruments:
LSI, a digital game-based learning attitude instrument, and fun-flow experiences to study how participants re-
act to it. It was found that the users who use their mobile phones had a positive attitude regarding game-based
learning. These learners have better communication between their peers which grant problems solving skills,
faster thinking and better education. However, they need to be taking into account the knowledge about indi-
vidual differences to be linking to the design process. The game based learning environment, it is not just a fun
experience, but it should complement the diverse range of learning styles.
Figure 3: The four ADOPTA playing styles together with the learning styles of Honey and Mumford and of
Kolb [3].
Notably, a question arises of whether there exists a correlation between learning styles and playing styles to
the recognition of both of them. So, a group of researchers [2] developed a project named ADOPTA, outlined a
20
family of playing styles games (Competitor, Dreamer, Logician, and Strategist) based on Kolb’s experimental
learning theory. In Figure 3, it is presented an explanation of each one of them as follows:
• Competitor: Players related to the Activist learning style, who prefer shooting and action games focused
on the competition. They can take risks and have fleeting thought in deciding tactics in the game. Also,
they are characterized by being open-minded and perceptive.
• Dreamer: Players related to the Reflector learning style, who prefer playing roles in a fantasy world of
avatars and observe the game-play. These players listen to the arguments of others by social interaction
and negotiation.
• Logician: Players related to the Theorist learning style, who prefer to analyze and solved patterns prob-
lems. They have efficient spatial awareness and contextual thinking. Each decision is taken in a rational
form by following the game rules, facts assimilation, and details precisely.
• Strategist: Players related to the Pragmatist learning style, who solves difficult tasks to find the most
efficient solution. They search to plan a strategy to test a hypothesis and take decisions. These players
do not enjoy shooting without reason.
Hence, ADOPTA focused on individual gameplay preferences can be used in the courseware adaption,
applying these game playing styles along with the learning styles of Honey and Mumford. To the identification
of critical features in the ADOPTA game playing styles, there was employed two questionnaires: i) ADOPTA
PSQ, a survey consisted of 40 dichotomous questions divided into four groups of 10 questions, corresponding to
the four playing styles; and ii) Honey and Mumford LSQ with 40 dichotomous questions, to expect correlations
of ADOPTA playing styles with the Honey and Mumford’s learning styles.
The experiment was done in 2015 with 315 Bulgarian respondents’ students at Sofia University “St Kl.
Ohridski”, Technical University of Sofia, and Plovdiv University “P. Hilendarski”. It was used the two ques-
tionnaires LSQ and ADOPTA PSQ, and two other surveys to stored information about the students’ demogra-
phy, and gaming experience. It was found that the most preferred types of games were: strategy, serious games,
RPG and puzzles, and the results for playing styles are similar to learning styles as being that both style families
are based on Kolb’s experimental learning theory.
Mainly, they concluded that Competitors tend to like fighting and shooter games, Dreamers play more
puzzles than other gamers, Strategists prefer strategy games and RPG, and Dreamers, Logicians, and Strate-
gist appreciate playing serious games. So, there exists a strong correlation between ADOPTA playing styles
and Honey and Mumford’s learning styles in a reliable analysis. Though, the research must be improved by
increasing the sample size of respondents and add more items in the ADOPTA PSQ.
21
Several study cases have been performed using a set of questionnaires in papers or web-based learning.
Nevertheless, one obstacle is that the students are bored easily and likely to give answers randomly without
taking into consideration the meaning of the question, obtain it an inaccurate classification. But, as nowadays
children and teens played video games for a significant amount of time with a noticeable attraction of them.
A group of researchers [24] decided to develop an entertainment video game named “The Adventure Hero”.
It is based on ILS in two dimensions: (Visual/Verbal) and (Active/Reflective), where the game decisions are
retrieved from the students’ answers in the game, a tool to develop understanding and learning style recognition.
“The Adventure Hero” was structured in two phases corresponding to the learning styles dimensions. It
allows students enjoying in the interactive game story with different situations, and learning personal recom-
mendation to improve its performance. The learning style could be diagnosed by the collection of the selected
responses on a series of quests. The game was developed of 11 questions with two answers to choose. The
obtained results were supported by five experts in education and game design to improve the mechanics in
the video game. The excellent property of this game is that it uses the main character in the story named Mr.
Hero, representing the student. While the avatar is walking in different adventures, it needs to communicate
with several aspects to retrieve the answer from the quest, by a selection in one of the icon responses. After the
game was finished, the 11 quests items were analyzed to diagnose the learning style intensity. The experimental
results were based on 79 first-year undergraduate students by completing the original ILS and then playing the
video game. The data stored in the experiment were the responses, time spent on the original ILS, time spend
on the game, learning style results, and a point scale questionnaire for perceptions and acceptance towards the
game.
Besides, it was found that the engaging game story “The Adventure Hero” could be used as a tool for
any student struggling with learning at school or university, and as a new way to bring changes in teaching
methods. This game obtained high accuracy in the detection of two dimensions of learning styles as a reliable
tool for learning style measurements. Despite, there are certain limitations such as lack of covering other ILS
dimensions, and in-depth research in gender differences, behavior patterns, and post-learning performance.
Also, another study [25] proposed the learning style detection in the ILS perception dimension in a video
game by tracking specific metrics (results obtained, the time elapsed, and levels) related to the student per-
ception recognition in a puzzle game named “Equilibrium”. The educational environment was personalized
immediately after the style detection and the monitoring in the game allowed to update in the system. The
puzzle rules were related to the placement of some figures (balloons and weights) in a balance such that the
torque of the figures was zero. The torque was given by multiplying the weight of the figure by the distance
from the balance’s center to the cell in which it was placed.
The experiment was done with a group of 63 junior Computing Science students in the context of courses
22
such as Exploratory Programming, Artificial Intelligence, and Software Engineering. The students were asked
to play different games, being one of them, the game “Equilibrium”. Precision detection was high in analyzing
how the student interacted with games, the researchers were able to predict the students who prefer to solve
complex problems and adopt detailed information. But one disadvantage was the sensitive detection to students,
who played a few times. As a result, little information could be retrieved to provide an accurate classification.
In that sense, a researchers group proposed an innovate approach to recognize the playing styles based
on the Kolb’s experiential learning theory [3]. This approach provided the opportunity to predict the learning
style of the player by multiple linear regression taken into consideration metrics in the game (task result, task
efficiency, and task difficulty), and self-report by the ADOPTA PSQ. Tasks as shooting, puzzle solving, and
discovering can be applied in the Competitor, Dreamer and Logician style, during fast decision making and
strategic vision in the Strategist style. The game was created to detect the playing styles, and it was known as:
“Rush for Gold”. The tasks are shooting, discovering gold bars or puzzle solving applied in the Competitor,
Dreamer and Logician style, while planning and problem solving was used in the Strategist style. As well,
emotions were inferred using Electrical Design Automation (EDA) signal to record facial expressions. The
main goal of the game was automatic recognition of playing and learning styles, in an educational maze about
business management.
To the research validation, it was done the first trial with 34 volunteers with the game demonstration, game
sessions, ADOPTA 40 items PSQ, and a post-game questionnaire. As a result, the research found that there
was a strong correlation of playing styles reported by ADOPTA PSQ with learning styles measured by Honey
and Mumford’s LSQ. The method of linear regression had higher accuracy than structured interviews. There-
fore, both methods were capable of not only recognize the ADOPTA playing styles but Honey and Mumford’s
learning styles. Although, improvement is to estimate another family of playing styles with other explanatory
variables, and to enhance the performance metrics of the game by using the emotional state. A summary re-
lated to the literature review about the detection of learning styles using video games is in the following Table 2.
23
Methods Main Application Pros Cons
Game based
on learning
to mobile
users, [23],
2006
Investigation of
Learning Style
Differences and
Attitudes toward
Digital Game-
based Learning
among Mobile
Users
Positive attitude toward
the digital game-based
learning
Lack in accommodating
learning styles
Mathematical
digital
games and
paper exer-
cises, [21],
2014
Comparing the ef-
fects of a math-
ematics game and
paper exercises
Significant correlation be-
tween the gains in the
working memory scores
and the changes in the
time required to complete
a computer mathematics
test
Enjoyment scores were
not significantly corre-
lated with the gains in the
cognitive measures
Video
Games and
Naive Bayes
Classifier,
[25], 2014
Detecting students’
perception style by
using games and
Naive Bayes Clas-
sifier
High precision in the per-
ception style of the stu-
dent
Sensitive detection to stu-
dents who played few
times
Serious
games in
mathemat-
ics, [22],
2016
Enhancing young
children’s math-
ematics skills
using serious game
approach
Help students improve
their cognitive level and
accelerate the learning
process
The game is not an auto-
matic diagnostic system
24
Methods Main Application Pros Cons
Story-based
Mobile
Game, [24],
2017
Diagnosing Learn-
ing Style and
Learning Sugges-
tion based on ILS
with an interac-
tive Story-based
Mobile Game
Application
High-level recognition on
two dimensions of learn-
ing styles
Cover just two ILS di-
mensions
Playing and
learning
styles ques-
tionnaires,
[2], 2018
Recognizing play-
ing styles based on
experiential learn-
ing theory
Strong correlation be-
tween ADOPTA playing
styles and Honey and
Mumford’s learning
styles in a reliable
analysis
Small sample size of re-
spondents
Adaptive
video game,
[3], 2018
Recognition of
playing and learn-
ing styles
High accuracy automatic
recognition of both play-
ing and learning styles
based on the Kolb’s the-
ory of experiential learn-
ing
Estimation of only one
family of playing styles
Table 2: Summary Table about video games to learning styles detection.
2.4 Predictive Learning Styles Detecting Models
Learning styles play a critical role to determine an efficient learning environment. In the traditional model has
been done by questionnaires, although they have been reliable in the detection. There is a problem, the lack
of enthusiasm by the students. For this reason, new approaches had emerged, such as video games, a potential
tool for learning style detection. It has an advantage in comparison with questionnaires since the information
to the detection is found by the student’s interaction with the system, instead of answering several questions on
25
paper. So, it has been used data-driven techniques that are more familiar with computer science researchers to
promote detection and improvement [10].
These techniques are Decision Trees, Genetic Algorithms, Neural Networks, Bayesian Networks, among
others. Decision trees are a classifier method able to be understandable by human beings. Two stages compose
this technique: building and pruning, being the most common algorithms for learning style recognition ID3,
C4.5, J48, and NBTree. At the same time, genetic algorithms can detect the learning style where each gen rep-
resents a student action in the system with new populations that show the student’s learning styles. Notably, the
most popular methods with accurate results have been ANN and BN. ANN is reliable in the speed of execution,
the ability to update extra parameters and generalize from specific examples. ANN approach often has been
presented with one input layer that contains the answers from questionnaires [26], interactions in a web system
[27], or data from Conversational Intelligent Tutoring System [15]; one hidden layer compounded of a differ-
ent number of neurons; and the output layer represented by learning styles based on Honey and Mumford, or
Felder and Silverman theories [28]. Meanwhile, BN is required as natural representation in a probabilistic way
with efficiently in encoding uncertain expert knowledge to learning style detection. This approach commonly
has been compounded by specific characteristics in the parent node; and in the children node, the learning
styles. BN have used the interactions in a Web-based education system to track for instance: how much time
the student takes to review a quiz or to finish an assignment [29], [30]. Also, BN can provide with an initial
information adaption related to the learning style and students’ preferences by the user interactions [31]. These
are the reasons that in this work have been used, these two predictive learning styles detecting models [10].
2.4.1 Artificial Neural Networks (ANN)
Artificial Neural Networks are computational models that simulate the neurons in the brain and had been applied
in the learning style recognition. The input layer is used to track the student behavior, the hidden layer to
the processing, and the output layer to the learning style detection. One approach proposed was the Fuzzy
Cognitive Maps (FCM) [32], a combination of fuzzy logic and neural networks techniques according to the
Kolb classification of learning styles. The inner layer takes the four learner profiles, while the output represents
the measurable learning activities to be diagnosed by the machine. Through this development, the researchers
were able to obtain a procedure to the learner’s profile and afterward provided to the student with its custom
material. By the machine appropriate diagnostic tests, it produced membership in fuzzy sets to the learning
style recognition. One disadvantage was human experts dependency and learner’s responses.
So, learning styles are essential in educational hypermedia systems. Hence, a group of researchers [27] used
a scheme composed by browsing behavior in a web-based educational system. The input layer represented the
26
learners’ browsing behavior. The factors used were: usage of embedded support devices, link types selection
and navigation between visited/unvisited nodes; while the number of outputs is equivalent to the number of
learning styles. The system was tested by answered 36 items from a questionnaire on 121 undergraduate stu-
dents majoring in Information Management of Chung-Hua University, Taiwan. As a result, the researchers were
capable of identifying the learning styles by using ANN with data extracted from the web-system. However,
one disadvantage was the small sample and the target population.
Even, that approach could be used in an adaptive user interface for e-Learning by learning style recognition.
For instance, a group of researchers [33] adopted ANN to recognize the learning styles in the users. The data
was gathered from the users’ logs during website navigation to find potential patterns in learning style acqui-
sition. The data was captured and stored in a database called Learning Repository, used in the Classification
Manager to provide with the appropriate user interface. The outcome was an automatic mechanism for style
recognition to the students.
At the same time, another study [15] used ANN to learning style prediction by using data from a Con-
versational Intelligent Tutoring System (CITS). The CITS is named Oscar, a live online system that tried to
emulate a human tutor by natural language tutorial. CITS predicts the learning style based on a set of rules
based on the Felder-Silverman model. However, a more robust model could be used depending on the inter-
actions with the system. The students’ answers and behaviors in the platform were captured, along with the
score given by the CITS. Therefore, they found specific attributes to put in the ANN related to the Processing
and Understanding dimension from Felder and Silverman. They obtained a high accuracy in the percentage
learning style detection, but further research to the other dimensions must be done to recognize the individual’s
learning style. Also, in [34], it was used the ANN technique to the identification of learning styles based on
the Felder-Silverman model using the four dimensions. The inputs were behavior data in a university course,
and the target was the learning style identified with the ILS questionnaire. They obtained the data from 127
computer science undergraduate students and found a high precision in learning style detection. This technique
can be applied as information to teachers to improve the learning style information to their students.
2.4.2 Bayesian Networks (BN)
Bayesian Networks have been applied to learning styles detection in the network precision to classify students.
BN represents a particular probability distribution and is associated with a set of conditional probability tables
(CPT) and nodes that represent learning style with specific characteristics. In [29], features were extracted
from the Web-based education system, with the relationship between each one of the nodes. The values of the
probabilities and CPT were obtained from the expert knowledge and experimental results with a mathematical
27
model based on Bayes’ theorem. For instance, factors analyzed as: whether the students revise the exams, and
how long the revision takes if it is considered the perception of the student. The probabilities were updated
parameters based on student behavior until it reaches an equilibrium. Therefore, it was evaluated 27 users in an
Artificial Intelligence course taken by Computer Science Engineering students to their learning style detection.
As an outcome, it was found a high precision in determining the perception style of the student. But, there were
mismatches in the understanding and processing dimension by the small sample size.
As well, Bayesian Networks can be used in expert systems to detect the learning style of the students.
In [35], the students answered a questionnaire, and they choose the answer in a randomly way. It takes im-
provement in choosing BN on an expert system that takes under consideration learner’s responses to the LSI
questionnaire to the classification of learners. The model was built on a set of learning styles corresponding
to a class, with a set of user responses in a binary way “yes” or “no”. Meanwhile, the conditional probability
was equivalent to the ratio of users who answers the questionnaire and after classified to a class. There was a
training of the BN by a classification via the LSI questionnaire. Then the application was tested, although it
was not used with real users.
Besides, a new Bayes model [31] was developed to adapt the initial information based on the learning style
and students’ preferences by observing the user system interactions. It was used as a learning style model and
decision model. The learning style model was represented with a BN, based on Felder and Silverman ILS.
In the decision model, was a BN Classifier to provide the right material to the student. It was composed by
selecting the appropriate objects to the learning style according to the characteristics of the students, prediction
of learning objective by BN, and the adequate knowledge to enhance the learning goals. A list of variables
represents the learning style with possible values such as the Input dimension equivalent to visual or verbal.
The model was able to obtain the recommendable learning object and the learning styles characterization of the
students. At the same time, generated data was managed to verify the classification of the initial model.
At the same time, to promote a better prediction in the learning style of students. A study [30] tackled
an uncertainty model developed in the LMS Moodle based on Felder-Silverman Learning Styles and Bayesian
Networks. The research was done with two groups of students from the Universidad Nacional de Loja and the
Universidad Internacional del Ecuador. It was established the Bayesian Network modeling in the virtual system
according to the resources and activities in the web-system. Being the parent nodes the web system interaction
and the children nodes to the learning style dimension such as perception or input. It was created a course related
to the “Bayesian networks introduction” to test the interaction in the system with 27 Ecuadorean undergraduate
students. Finally, the study allowed to predict the students’ learning styles by the Felder-Silverman theory.
One task that can be improved is the definition of the resources and activities variables in the virtual learning
environment in LMS Moodle. A summary related to the literature review about the automatic learning styles
28
detection using Artificial Neural Networks and Bayesian Networks is in the following Table 3.
Methods Main Application Pros Cons
Combination
of Fuzzy
Logic and
Neural
networks
techniques,
[32], 2004
Learner’s Style and
Profile Recognition
via Fuzzy Cogni-
tive Map according
to Kolb classifica-
tion
Direct use of machine
diagnostic test to the
learner’s style recognition
Dependence on human
experts and the learner’s
responses
Multi-layer
feed for-
ward, [27],
2005
Learning styles
recognition of
learners by ob-
serving browsing
behavior through a
neural network
Learning style identifica-
tion with an acceptable
level
Small sample size and
tested in college students
Bayesian
Networks,
[29], 2007
Detecting students’
learning styles by
using Bayesian
Networks in the
Felder and Silver-
man learning styles
model
High precision in de-
termining the perception
style of the student
Mismatches were found
in the understanding and
processing dimension
Expert sys-
tem based
on Bayesian
Networks,
[35], 2007
Learning styles’
estimation using
Bayesian Networks
New Bayesian Network
classification model to
learning styles
No tested with real learn-
ers.
29
Methods Main Application Pros Cons
Bayesian
Model, [31],
2009
Update in the
learning styles
based on the stu-
dent Bayesian
model and its
preferences
Adaptive learning envi-
ronment applying learn-
ing styles
No tested with real users,
generated data
An artificial
neural net-
work with
an adaptive
user inter-
face, [33],
2010
Learning style
identification using
Artificial Neu-
ral Network and
Felder and Sil-
verman Learning
Styles for adaptive
user interface in
e-learning
Automatic mechanism for
style recognition
No information about the
users that test the systems
Multilayer
Perceptron
Artificial
Neural Net-
work with
a conver-
sational
intelligent
tutoring
system, [15],
2013
Profiling learning
styles from a con-
versational tutorial
using a multilayer
perceptron in the
Felder and Sil-
verman learning
model
Test with real captured
data with the conversa-
tional tutorial
Used only two Felder Sil-
verman dimensions
30
Methods Main Application Pros Cons
Bayesian
Networks,
[30], 2014
Students Learning
styles’ prediction
using Bayesian
Networks in virtual
environments
Efficient accuracy Improvement in the con-
ditional probabilities ta-
bles
Artificial
intelligence
models to
automatic
learning
style recog-
nition, [10],
2015
Literature review
about methods
to allow auto-
matic detection of
learning styles
No available No available
Artificial
Neural Net-
works, [34],
2015
Learning styles
identification using
Artificial Neural
Networks in the
Felder and Silver-
man learning styles
model
Efficient accuracy in arti-
ficial neural network ap-
proach to identify student
learning styles
Small sample size and tar-
get population
Table 3: Summary Table about Artificial Neural Networks and Bayesian Networks to learning styles detection.
31
3 Techniques for Learning Style Detection
The majority of techniques used to detect learning style recognition have been static methods such as question-
naires, and in most of the cases, it has been applied to undergraduate students. However, children in their early
years in the schools are the primary population to future career development. In that sense, it was chosen as a
static method, the CHAEA-JQ, by its reliability in educators and academic counselors to test children [20].
Certainly, CHAEA-JQ has been a technique used in decades by has been a tool that produces boredom in the
student assessment. For this reason, to provide entertainment in the evaluation of learning style identification,
in recent years, has been used video games. Being the reason that in this work, it has been developed video
games to recognize the learning style focused on ADOPTA playing styles based on Kolb work [36].
The previous techniques are non-predictive ones, in that sense, new research in Artificial Intelligence tech-
niques for learning styles detection have emerged to provide automatic learning style recognition. ANN [37]
was chosen as being recognized as a reliable technique in the speed of execution and the updating of parame-
ters. Such that the input in the network are variables related to the characteristics of the students based on their
questionnaire answers, and the output is represented as the percentages from the game’s score. Also, to obtain
the relation between certain variables such as parent nodes (Mini-Games and Answer Time) with the children
node (Learning Styles), it was chosen BN [29] to represent the relation between them. So, four techniques, two
non-predictive (CHAEA-JQ, video games), and predictive ones (ANN, BN) have been selected to gathering
data to recognize the students learning style.
This chapter presents a brief overview of the methodology used in this work, and next, it is divided into
four subsections: CHAEA-JQ, mathematical Mini-Games, ANN, and BN.
3.1 Methodology
The general process in this work can be seen in Fig 4. The first step was the research about related works that
detect students learning styles with the use of questionnaires, video games, and artificial intelligence techniques.
Then, it was targeted a sample of 100 students from primary level education level from the school “Teodoro
Gomez de la Torre” (Ibarra-Ecuador). Besides, it was started the development of the tools used in this research
that will explain after on, and finally, after the student was tested with the different learning styles models, it is
possible to detect the learning style of each student.
32
Figure 4: General block steps to detect the learning styles.
A detail explanation about the learning styles models detection can be seen in Fig. 5. After the imple-
mentation of the CHAEA-JQ in the web-system, mathematical Mini-Games in C# programming language, BP
algorithm in Matlab and BN in C# programming language. It was done the first trial, where the students an-
swered the CHAEA-JQ and played the mathematical Mini-Games to gather data to be applied in the BP. In the
second trial, the CPT tables were created based on the data obtained before. Then the students played the math-
ematical Mini-Games where it was included the BN algorithm. To finally, with the four techniques obtained
the learning styles percentages in the testing sample from the students.
33
Figure 5: Detail steps to detect the learning styles.
3.2 Methods
3.2.1 CHAEA-Junior Questionnaire (CHAEA-JQ)
CHAEA-Junior is a questionnaire to students in the elementary education level and first years of high school.
It is based on Honey and Mumford learning model. The test identifies the learning styles preference of the
students (output) by a set of questions written according to the psychological children characteristics (input).
It is characterized by its usability, and speed, both in its application and in its correction by counselors and
teachers.
The standard questionnaire is presented in a single folio sheet consisted of 44 questions, distributed ran-
domly, with four groups of 11 items corresponding to the four learning styles: Activist, Reflector, Theorist, and
Pragmatist. The absolute score obtained in each style is a maximum of 11, showing the level reached in each of
the four Learning Styles. The student needs to answer by drawing a circle in the item that he/she agreed. Oth-
erwise, it can leave the item without surrounding. On the back of the folio, four columns of numbers belonging
to each of the four Learning Styles are presented to define the student’s preferred learning profile.
34
Figure 6: CHAEA-Junior Block Diagram.
In Fig. 6 shows that to this research it was selected randomly 24 questions from the standard question-
naire due to time constraints in the experiment, and to avoid boredom in children. Each group of 6 questions
represents a learning style, and it was not used a folio sheet as the standard version. Instead, the questions
were written in a web system to this research purposes. The questionnaire was developed using Unity 3D, C#
programming, and MySQL database, as seen in Fig. 7 to a better description of the client/server architecture. In
Fig. 8 there is an example from the web-system. It is composed of interactive questions, entertainment music,
an animated green avatar in the right-bottom screen, colorful background, and clear font size. A “next button”
was added that is activated after 5 seconds to verify that the student has a recommendable time to answer the
twenty-four questions.
The student needs to check in a check-box if he/she is agreed with that question, this will be equivalent to
increase the score in that corresponding learning style. Otherwise, if he/she disagrees, the check-box will be
empty and equivalent to no increase in the score in that learning style. Then, these answers are represented in
a binary way, 1 (agreed) and 0 (no agreed) to apply a binary sum to recognize the learning style of the student.
35
(a) Architecture Diagram (b) Deployment Diagram
Figure 7: (a) Architecture and (b) Deployment Diagrams in the Web Application.
Figure 8: Screenshot from the CHAEA-Junior Questionnaire.
3.2.2 Mathematical Mini-Games
This method is compounded by four mathematical Mini-Games. That identifies as output the learning styles
percentage of the students. This method is composed of a set of 96 mathematical questions, being 24 questions
36
designed according to each Mini-Game playing style. The input is represented by the game scoring achieved
in each mathematical Mini-Game at the first playing session.
For this work, it was developed 2D mathematical Mini-Games using the Unity 3D engine, C] programming
language, PHP, MySQL, and sprites with different colors representing the avatars in the game. The Mini-Games
were embedded in a Web Application, the software architecture can be seen in Fig. 7. The objective is to solve
basic mathematical operations: sum, subtraction, multiplication, and division. Each mathematical operation is
composed of two numbers and the symbol to represent that operation. Taken into account the topics in the book
from the “Ministerio de Educacion del Ecuador en Matematicas”, and advice from different teachers from the
school “Teodoro Gomez de la Torre” (Ibarra-Ecuador). The Mini-Games are integrated by three levels, in the
following way:
• Level 1: number 1 (1 - 30), number 2 (1 - 15)
• Level 2: number 1 (31 - 60), number 2 (16 - 30)
• Level 3: number 1 (61 - 100), number 2 (31 - 50)
Each Mini-Game was designed based on ADOPTA playing styles based on Kolb’s learning styles. At the
same time, each one of them has an interactive and friendly user interface for the children with specific colors
and engaging music. The student played four mathematical Mini-Games (Competitor, Dreamer, Logician, and
Strategist).
The Competitive game as seen in Fig. 9 has the mathematical question in the upper part of the screen with
a spaceship avatar. The avatar shoots to the correct answer between three possible answers, at this moment is
heard a sound with the rocket exploiting.
37
Figure 9: Screenshot from the Competitive Mini-Game Style.
In the Dreamer game as seen in Fig. 10 it was developed a puzzle game, the mathematical operation is in
the middle of the screen with a missing piece. The answer to the mathematical operation must be chosen from
the three possible answers on the right side of the screen. Also, there is an interactive green avatar that opens
and closes the mouth to mimic an avatar talking being a characteristic that attracts the Dreamer style.
38
Figure 10: Screenshot from the Dreamer Mini-Game Style.
As seen in Fig. 11 the Logician game was designed with cards games. At the screen center, there is the
mathematical operation with two numbers, and in the bottom, there are three possible answers. The Logician
has a preference to answer in a step by step process reason, so the student has to see the question and then press
the correct answer.
39
Figure 11: Screenshot from the Logician Mini-Game Style.
In Fig. 12 is shown the Strategist game composed in an RPG format, first-person video games with obsta-
cles. In this sense, the mathematical operation is in the upper part of the screen, and the green avatar must walk
and jump to shoot bubbles to the correct answer. The three possible answers are in different blocks to challenge
the student.
40
Figure 12: Screenshot from the Strategist Mini-Game Style.
3.2.3 Artificial Neural Networks (ANN)
This work proposes an Artificial Neural Network design compounded by three kinds of layers: an input layer,
hidden layers, and output layer as it is depicted in Fig. ??. Where the input layer contains the neurons that
receive as entry data the CHAEA-Junior answers of the 24 questions from the questionnaire, and the target are
the percentage of each learning style obtained from the mathematical Mini-Games based on ADOPTA styles.
To calculated this percentage, after the student answer the questions in each Mini-Game, it was measured the
percentage of each one by the division between the score obtained in each Mini-Game divided by the maximum
score between them, which have been applied to the training students’ group. The output layer contains the
neurons that provide the prediction of the student learning style which is correlated with ADOPTA styles.
The network learning process takes the input neurons and the expected output neurons, in order to update the
weights on the internal neurons of the one hidden layer layers until getting the most likely computed output
neurons with respect to the target. The network learning process uses the BP algorithm in order to propagate
back the derivative of the error function from the end to the start of the network. The difference between the
result from the target and the output of the BP is used as a back propagation error.
41
Figure 13: Top-level architecture of the artificial neural network approach.
The BP main equation is given by equation 1, where oi is the output of the neuron belonging from the
hidden ni, p represents the synaptic potential, wij are the synaptic weights between neuron i in the current
layer and the neurons of the previous layer with activation oj . Therefore, the sigmoid activation function is
computed as shown in equation 2.
oi = s(n∑j=1
wij · oj) = s(p) (1)
s(x) =1
1 + e−βx(2)
The BP algorithm objective is to reduce the error obtained by modifying the synaptic weights, to get a
minimum difference between targets and network outputs. The error is given by equation 3, where the first sum
is computed on the p patterns of the data set and the second sum is calculated on the N output neurons. ti(r) is
the target value for output neuron i for pattern r, and oi(r) is the response network output.
E =1
2
p∑r=1
N∑t=1
(ti(r)− oi(r))2 (3)
42
The synaptic weights between two last layers of neurons are given by 4, where η is the learning rate and
s′ is the derivative of the sigmoid function oi, and the other weights are modified according to deltas (δ) that
propagate the error.
4 wij(r) = −η∂E
∂wij(r)= η[ti(r)− oi(r)]s′i(pi)oj(r) (4)
For this work, code routines in Matlab were developed for the calculations of the neural network output,
deltas, weight updated, and transfer function. The code was implemented based on the standard formulas
from the BP algorithm. Data from the database of the binary CHAEA-JQ answers, and the scoring of the
mathematical mini-games was saved in a .csv format file. This data was used in the BP algorithm to the
training and testing procedures for the algorithm. The implementation is hosted on [38].
3.2.4 Bayesian Networks (BN)
This work proposes a Bayesian Networks design composed of nodes and arcs that represent the relationship
between them, with parent nodes and a children node. The parent nodes are the Mini-Games and Answer Time
(input), while the children node represents the Honey and Mumford Learning Styles (output).
In this sense, a BN model shows the relationship between the learning styles with the respective features.
The BN is shown in Fig. 14, where the parent node known as Mini-Games has four states: Competitive,
Dreamer, Logician, and Strategist based on ADOPTA playing styles. In the initial state has a value of 25%
each one of them. The values are changed depending on the answers of the mathematical questions in the
Mini-Games. The parent node named Answer Time has two states: High (0-6.5 seconds), and Low (6.6 - 13
seconds). Each one has a value of 50% at the beginning phase. Then, the children node denominated Learning
Styles has all the possible combinations of learning styles values: Activist, Theorist, Pragmatist, Reflector,
Activist/Reflector, Activist/Theorist, etc. At the initial stage, each one of them starts with 0% and it increases
according to the interactions between the student and the web-based games system by analysis of the answers
from the mathematical questions using the different Mini-Games and the Answer Time that took to resolve them
with the various levels. After, a certain amount of time, the nodes are updated as is shown in Fig. 15. Each
parent node has a CPT table calculated from expert knowledge (CHAEA-JQ) by identifying qualitative problem
aspects, such as direct relationships between the problem variables, and experimental results. CHAEA-JQ and
the results obtained from the interaction with the system are used in the training phase respectively.
43
Figure 14: Initial state Bayesian Network diagram.
44
Figure 15: Bayesian Network updated after the student answers the question.
Bayesian Networks are based on the Bayesian rule P (H|E) = P (E|H)P (H)P (E) , where:
• H is a probability variable that denotes a hypothesis existing before evidence.
• E is a probability variable that expresses observed evidence.
• P(H) is the prior probability, the hypothesis’ initial value.
• P(H | E), represents the conditional probability of H given E, which is called posterior probability. P(E
| H) is the conditional probability of occurring evidence E when the hypothesis is true. Where, the
likelihood ratio is P(E | H) / P(E), but P(E) is a constant value. So, it can be considered P(E | H) as a
likelihood function of H with an E fixed.
• P(E) is the probability of occurring evidence E with all mutually exclusive hypotheses cases.
45
In a general setting for probabilistic inference is defined a set L of propositional variablesL = {L1, L2, ..., LN}.
The evidence are the variables in a subset E of L which have certain definite values, E=e (true or false). To
calculate the conditional probability, P(Li=li |E = e), it is defined that any variable Li has value li, given the
evidence. This process is called probabilistic inference. Since Li has the value true or false, there are two
conditional probabilities, P(Li = true | E=e) and P(Li = false | E=e). Using the definition for conditional
probability, it is found:
P (Li = true|E = e) = P (Li=true,E=e)P (E=e)
P (Li = true|E = e) is obtained by using the rule for calculating joint probabilities:
P (Li = true,E = e) =∑
Li=true,E=e P (L1, ..., Lk)
So the marginal probability values of the learning style node are found with the values of independent
nodes. For example, if the student has the learning style activist as the dominant learning style, the probability
is P(LS=Activist |Mini-Games, Answer Time). The general form in the equation can be formulated as:
PLSBN = p(LS = K |MG,AT ) = p(LS = K |MG = C,AT = H)p(MG = C)p(AT = H) +
p(LS=K
|MG = C,AT = L)p(MG = C)p(AT = L) + p(LS = K |MG = L,AT = H)p(MG = L)P (AT =
H)+p(LS = K |MG = L,AT = L)p(MG = L)p(AT = L)+p(LS = K |MG = S,AT = H)p(MG =
S)p(AT = H) + p(LS = K |MG = S,AT = L)p(MG = S)p(AT = L) + p(LS = K |MG = D,AT =
H)p(MG = D)p(AT = H) + p(LS = K |MG = D,AT = L)p(MG = D)p(AT = L)(5)
The variables can be represented as LS (Learning style), Mini-Games (MG), Answer Time (AT), Compet-
itive (C), Logician (L), Strategist (S), Dreamer (D), High (H), Low (L); and where K could be replaced by
Activist, Theorist, Pragmatist, Reflector, or any combination of each learning style and the web-based games
system.
The proposed algorithm calculates the learning styles probabilities, taken into account the prior Mini-Game
probability and prior Answer Time probability. Each one initializes on a certain value, and the posterior learn-
ing style probability is initialized in zero. There are an increment and decrement of an established quantity
according to the answers applied of the student to update the percentage in the Mini-Games and Answer Time.
Finally, after answering questions, the posterior learning style probability is updated and the level of the video
46
game is increased. The game session with BN was twenty-one minutes, where the values of the nodes are
updated.
This algorithm was developed in C# programming, and embedded in the Mini-Games based on ADOPTA
playing styles. Where the posterior probability was updated based on the parent nodes. The implementation is
hosted on [38].
47
48
3.3 Experiment Set-up
In the experimental set-up were considered certain conditions, availability in the primary school, time in data
recollection, and resources for this project. So, this subsection is divided in the instruments used, participants,
data recollection and preparation to then be calculated by specific metrics.
3.3.1 Instruments
To identify the percentages of learning styles in each student, it was used the 24 questions related to the
CHAEA-JQ, and 24 mathematical questions based in each Mini-Game. The total game session was composed
of 96 mathematical questions of four basic operations (addition, subtraction, multiplication, and division). The
data from the CHAEA-JQ and the Mini-Games was used in the BP algorithm. Finally, a game session was ap-
plied to the students in 21 minutes. The school computer laboratory was composed of twenty DELL computer
machines, with 512GB, 8GB RAM and Operating System Windows 10.
3.3.2 Participants and procedure
The participants in the experiment were 100 students in seven-grade from the school “Teodoro Gomez de la
Torre”. The students were divided into groups of 20 students to be part of the experiment given by the laboratory
capacity. In the beginning, a presentation of 10 minutes was given to the students in order to prepare them to
understand the mechanics or questions in the Mini-Games. Then, it was necessary two trials, in the first trial,
the students took the interactive CHAEA-JQ in the web-system, and play the Mini-Games based on ADOPTA
playing styles. These data were used in the BP algorithm, and to train the BN. In the second trial, it was selected
20 students randomly to test the algorithm based on BN in a 21 minutes game session.
3.3.3 Data Preparation
Experiments were carried out over collected data of 100 students between 11 and 12 years old of the school
“Teodoro Gomez de la Torre” (Imbabura-Ecuador), which are divided randomly into two data sets, 80% for
training and 20% for testing phase. The four methods, CHAEA-JQ, Mini-Games, ANN, and BN, were evalu-
ated on the testing set.
3.3.4 Quality Metrics
The ability to identify the percentage of each learning styles that have a student was measured through met-
rics such percentage for learning style recognition for CHAEA-JQ (PLSJQ), Mini-Games (PLSMG), ANN
(PLSANN ) and BN (PLSBN ). PLSANN corresponds to the output of the network in equation 1, PLSJQ for
49
each learning style is given by equation 8, PLSMG for each learning style based on ADOPTA playing styles
is given by equation 11, and finally PLSBN it can be found with the equation 5. PLSJQ takes into account
the Summatory of Each Learning Style (SELSL), and the Summatory of all questions related to the Learning
Styles (SLSJQ).
SELSL =
6∑j=1
qLj (6)
where qLj = 1 if it is check the checkbox, and qLj = 0 if it is not check the checkbox.
SLSJQ =
24∑j=1
qj (7)
PLSJQ =SELSLSLSJQ
(8)
PLSMG is calculated with the Summatory of each Learning Style based on ADOPTA playing style (SELSADOPTA)
in each Mini-Game level, and the Summatory of all the scores related to the Learning style based on ADOPTA
playing style (SLSADOPTA).
SELSADOPTA =
3∑level=1
8∑m=1
qADOPTAlevel,m (9)
where qADOPTAlevel,m = 1 if it chosen the correct answer in the Mini-Game, and qADOPTAlevel,m = 0 if it is chosen
the incorrect answer in the Mini-Game.
SLSADOPTA =
96∑j=1
qj (10)
PLSMG =SELSADOPTASLSADOPTA
(11)
50
3.3.5 Experiment Description
It is assessed the quality metrics achieved by the four analyzed methods. Several experimental tests have been
performed in a group of children by answer the questions in the CHAEA-JQ and by answering mathematical
questions in the Mini-Games. An example of one of the questions to the Theoretical learning style is:
� Estoy seguro de lo que es bueno y lo que es malo, lo que esta bien y lo que esta mal.
To test CHAEA-JQ, each student answered 24 questions in an interactive web-system. The answers are
captured in the system, with 1 corresponding to agree (checked checkbox) and 0 correspondings to disagree
(empty/unchecked checkbox). An example of data captured in one student will be:
101 · · · 11111
where each number represent the answer of the 24 questions. By using (PLSL), it can be found the
percentage of learning style with the CHAEA-JQ. An example of one student will be:
0.00 | 0.50 | 0.17 | 0.33
where each value represents learning styles percentage in the questionnaire, like 0.00 it means the percent-
age of the Activist learning style, meanwhile 0.50 represents the Reflector learning style. Then, the student
played four mathematical Mini-Games (Competitor, Dreamer, Logician, and Strategist). The goal of the Mini-
Games is to solve basic mathematical operations according to the rules in each one. If the student answers
the question correctly, the score is increasing at one point in that Mini-Game, otherwise, there is not a score
increment. After the student finished playing the four Mini-Games, the data was captured in the system. An
example of the scoring results of one student will be:
120 | 100 | 130 | 200
Each value in the expression represents the score in one of the Mini-Games. For instance, 120 represent the
score of the Competitive style; meanwhile 200 represent the score of the Strategist style. Then, it is calculated
the percentage of the learning style based on the scores from the Mini-Games. The percentage is calculated by
dividing the score of each game by the sum of the scores of all the Mini-Games, obtained as a result:
51
0.22 | 0.18 | 0.24 | 0.36
Therefore, the binary answers from the CHAEA questionnaire were used as input in the BP, and the tar-
get was the percentage calculated from the mathematical Mini-Games based on ADOPTA playing styles, as
explained before.
For the training and testing processes, ANN uses η = 0.05 and β = 1/2, using 1000 for the maximum
number of iterations as the parameters that best fit the model in different experimental tests. Besides, different
architectures were utilized with the calculation of the Mean Square Error (MSE), the best performance with
one hidden layer of 10 neurons.
The BN adaptive algorithm takes as nodes each Mini-Games and Answer Time with edges that are con-
nected to the Learning Styles node. The data collected of 80 students will be used in the training phase to build
the CPT learning styles tables based on the students’ answers with 24 questions related to the CHAE-JQ, and
the Mini-Games results.
Around 20 students played the Mini-Games using the adaptive algorithm with BN after the game session is
finished, the system detected the posterior learning styles probability percentages of each student. Algorithm 1
depicts how as the student is playing the game the posterior learning probabilities are updating (UpdateProba-
bilitiesBN), that is calculated similarly by the equation 5. An example of one student will be:
0.18 | 0.19 | 0.11 | 0.03 | 0.00 | 0.02 | 0.00 | 0.10 | 0.02 | 0.06 | 0.02 | 0.00 | 0.25 | 0.02 | 0.00
In the previous expression, the posterior probability of each combination of the student learning style is
calculated. For instance, the value 0.18 represents the posterior probability of being an Activist given the Mini-
Games scores and the Answer Time p(LS=Activist |Mini-Games, Answer Time). Meanwhile, 0.02 in the fifth
position represents the percentage of being an Activist/Theorist. However, in this study is going to be used the
four posterior probabilities from the learning styles: Activist, Reflector, Theorist and Pragmatist by weighted
the sample to be one hundred percent.
In that way, the four techniques were evaluated on the testing set, finding the error between the percentages
in the learning styles calculations and comparing with the CHAEA-JQ. Therefore, the error (∈) between the
CHAEA-JQ and other technique for the four Learning Styles (N=4) was calculated by the equation 12.
∈=
∑N1
abs(PlsX−PlsJQ)PlsJQ
N(12)
52
where X can be represented with the values of Artificial Neural Networks, Mini-Games or Bayesian Net-
works.
53
4 Results, Discussion and Conclusion
The results achieved by CHAEA-JQ, Mini-Games, ANN, and BN for learning style identification are presented.
This chapter is distributed in three sections: results and discussion, restrictions for high precision in learning
style identification, and conclusions.
4.1 Results and Discussion
Each technique is compared concerning the CHAEA-JQ to evaluate the precision reachable to recognize the
learning styles, which during decades has been considered as the most reliable by teachers and counselors.
The learning styles percentages in the student testing group were calculated by PlsJQ, PlsMG, PlsANN and
PlsBN , then the percentages were averaged as shown in Table 4, where the first column represents the learning
styles. In the second column is described the percentage of learning style from the CHAEA-JQ, in the third
column the percentage calculated from the scores in the Mini-Games, the fourth column represents the output
in ANN, and in the last column the percentage of learning styles based on the algorithm using BN.
Table 4: Average of percentages learning styles based on CHAEA-JQ, Mini-Games, Artificial Neural Networks,
and Bayesian Networks.
Learning Styles PlsJQ PlsMG PlsANN PlsBN
Activist 18.82% 24.29% 23.44% 34.67%
Reflector 32.45% 31.24% 29.91% 37.76%
Theorist 22.33% 24.15% 26.70% 21.45%
Pragmatist 26.39% 20.33% 20.07% 6.13%
According to Table 4 taking the CHAEA-JQ as the most precise method in learning style. It is shown
that from the four learning styles models, the higher percentage in learning style is the Reflector, followed by
the Pragmatist, Theorist, and Activist. The Reflector learning style states that most of the students learn in
processing information by thinking before any further decision. This type of learning style adapts in how the
students learn in the classroom with the traditional educational system. The Reflector is a learning style, where
the student tries to obtain conclusions by seeing the problem in different perspectives, and interact with an
animated green avatar as a fun activity in this research. At the same time, the Theorist fits in third place in the
54
percentage recognition between the CHAEA-JQ, Mini-Games, and BN. Being that this type of student acquires
knowledge in a step by step process.
In order to determine how much the identification given by the evaluated models varies. It is essential
to mention that the results found in the CHAEA-JQ have a standard deviation of 5.85%. The variation in
this technique states a considerable difference in the dispersion between the data. The reason is the highest
percentage found in the Reflector learning style with a value of 32.45% in comparison with the other learning
styles. This predominance in this learning style is due to the psychological evolution in children by following a
traditional educational system, where the students need to adapt to that learning style since their early stages of
life. Being this tendency to this learning style to listen first, then act to conclude, and to create solutions; with
a tendency to be thoughtful and cautious. The students with a higher-level Reflector learning percentage do not
learn when they are forced to take a leadership position in a group and doing tasks without prior preparation.
The Mini-Games have a standard deviation of 4.55% as it can be seen in Table 4 with the lowest value in
the Pragmatist learning style and the highest value in the Reflector learning style. There is no higher amount
of dispersion of data due to how the data is ordered. The ANN approach has a standard deviation of 4.23%,
meaning that the spread of the information is more uniform, as it is depicted in Table 4 starting from the lowest
percentage style which is 20.07% in Pragmatism learning style until the most significant value with 29.91%
in the Reflector learning style. The BN has the highest standard deviation of 14.43% among the three other
methods due to the 6.13% found in the Pragmatist learning style, and the Reflector as the dominant learning
style with a value of 37.76%.
55
0
10
20
30
40
50
60
29.9131.24 32.45
37.76
Learning Styles Detecting Models
Refl
ecto
rLea
rnin
gSt
yle
Perc
enta
ge
ANN Mini-Games CHAEA-JQ BN
Figure 16: Reflector Learning Style Percentage vs. Learning Styles Detecting Models.
Fig. 16 depicts information about the percentage found in the Reflector percentage with each one of the
techniques for learning style identification. In the x-axis are represented the Learning Styles Detecting Models,
and in the y-axis the Reflector Learning Style Percentage. The minimum percentage in the recognition is ANN
with a value of 29.91%, then it follows the Mini-Games with 31.24%, CHAEA-JQ with a measure of 32.45%,
and the highest result is achieved by BN with 37.76%.
The CHAEA-JQ will differ if the 24 questions were selected different from the 44 items and if the ques-
tionnaire is presented with graphical representation in each question. It will produce an increase or decrease
in the learning style percentage in the student. Although, the recommendable main improvement is boosting
interactivity in the questionnaire to amuse the students to answer the question self-aware. Because even that
the survey is the most common approach, the student could lie, producing an unreliable classification.
Meanwhile, in the Mini-Games, there were not similar results in the Activist and Pragmatist in comparison
with the CHAEA-JQ. It is due to the mechanics in the game, in the Activist style related to the Competitive
playing style, characterized by quick and risk thinking to shoot rockets to the correct answer. Some students
answered the questions randomly without trying to solve the item correctly. On the other hand, one the problem
with the Pragmatist style related to the Strategist playing style is that the students were confused in jumping
56
the avatar correctly to throw away bubbles to the correct answer. Where the primary goal to this learner is the
functionality of the game.
Likewise, ANN only fits in the Reflector style percentage in comparison with the CHAEA-JQ. It happens
because in the Theorist related to the Logician playing style, the students were confused with pressing the card
center in the correct answer selection. As an outcome some questions could not be answered by the students.
By fixing some components in the game, the data gathering will be more effective in the prediction of learning
styles percentages.
Finally, when it was used the BN algorithm, the most prominent percentage was the Reflective style that
fits in the learning style recognition from the CHAEA-JQ, followed by the Theorist style. The other learning
styles do not correlate with the CHAEA-JQ. One of the reason in the Activist style it was by game mechanics.
Some students answer the question randomly at the moment the rockets were shooting. So, by the relationship
between the parent node Mini-Games, and children node Learning Style, according to the answer to the question
causes that the children node is updated. That being said, it is the cause that with the BN algorithm, the outcome
is that the Activist has the most significant percentage. Meanwhile, the Pragmatist style is related that the
students do not understand clearly how to jump and throw away a bubble to the correct answer in the game.
It is essential to mention that, the CHAEA-JQ is a static method, so it cannot be dynamic to adapt to
recognize the learning style of the student. Even it is not useful in providing a specific type of video game
related to ADOPTA playing styles because it is calculated by agreement or disagreement in the set of questions
answered by the students.
Instead, the ANN approach can be modified according to the input in the network, and even the capabilities
prediction can be faster and reliable to learning style identification based on playing styles with educational en-
tertainment games. The data collected from the CHAEA-JQ and the scoring from the mathematical mini-games
based on ADOPTA playing styles is recommendable to be used on research about learning style identification
based on Honey and Mumford theory. Also, BN played an essential role in learning style recognition, estab-
lishing nodes that have a relation between them in using data from Mini-Games and Answer Time; with a CPT
filled up with expert knowledge and experimental results.
57
0
10
20
30
40
50
60
15.9818.97
45.32
Learning Styles Detecting Models
Ove
rall
Err
orPe
rcen
tage
Mini-Games ANN BN
Figure 17: Overall Error Percentage vs. Learning Style Detecting Models.
The results depicted in Fig. 17 shows the error between the CHAEA-JQ comparing with Mini-Games,
ANN, BN. The better tool in the overall learning style recognition is the method that uses the Mini-Games
(blue bar) with an error of 15.98% characterized by its reliability and validity. Followed by ANN (red bar), an
alternative tool with an error of 18.97% using the architecture of one hidden layer and ten neurons as the best
performance; and BN (brown bar) with 45.32%.
The error in the Mini-Games is related to the mechanics involved in them, as explained before. The error
in ANN is by the amount of data that was collected because according to the input in the network and the
scoring in the Mini-Games, it will improve drastically the learning style percentage prediction. Data gathered
from the mechanics in the game, for instance, in the Activist style related to the Competitive playing style. The
students attracted by the game, shoot to an answer randomly without trying to solve the question correctly. On
the Pragmatist style related to the Strategist playing style, the students were confused in jumping the avatar
precisely to throw away bubbles to the correct answer. In the Theorist style, related to the Logician Style, the
students were confused with pressing the card center in the right answer selection. Those behaviors generated
that some students did not answer some questions. Also, the reliability in learning style detection can be
enhanced by sample size amplification because twenty students compounded the testing group in this work.
58
Furthermore, in the adaptive algorithm using BN, the detection has some inaccuracies. One of the reasons
to improve is increasing more parent nodes to the learning style detection, and updating parameters. Where the
posterior probability was calculated given the parent nodes are based on how many correct answers the student
had, and the time that took to answer them. At the same time, the selection of the decision in presenting the
mathematical question to the student. Besides, the algorithm implemented in this research was the first attempt
because in the majority of cases is used BN in web sites to tackle the interactions in the system. Interactions
such as how much time the student clicked a specific button in a website, or how much was the time that the
student took in read a paragraph related to a topic. But, using Mini-Games related to ADOPTA playing styles
with BN, have not been implemented in this context.
Notably, that the errors in the different Learning Styles Detecting Models will diminish if is considering cer-
tain characteristics for high precision learning style identification such as the number of questions, sample size,
game metrics, game types, student mental state, different games, game experience, demography, computational
resources, and environment.
4.2 Restrictions for high precision in learning style identification
This work was done in four stages: research, sample selection, development, and testing, as it was depicted in
Fig 4. The sample selected was students from seventh grade from the “Unidad Educativa Teodoro Gomez de
la Torre”, by administrative agreements and time constraint it was possible to recollect data from 100 students
in three weeks. Results found in Table 17 show the error in learning style detection in each method. However,
high precision can be accomplished if it is considered the following conditions:
• Number of questions: The students answered the CHAEA-JQ composed by twenty-four questions re-
lated to Honey and Mumford work. However, the reliability in learning style detection can be enhanced
by amplified the sample size as explained in [39]. He recommended that the number of questions must
be at least five times greater than the number of variables.
• Sample size: It was chosen 80% for training and 20% for testing in both artificial intelligence techniques
ANN and BNN. So, a prior statistical analysis is recommendable to an accurate sampling size. Besides,
the sample size tends to be small in the field of learning style automatic recognition in the majority of
studies [19],[24],[2],[3], there exists criticism related to its efficiency. An increment in the population
size will enhance the learning predictive capabilities and reliability in using the BP algorithm and BN.
• Game metrics: In the mathematical Mini-Games based on ADOPTA playing styles, can be improved by
selecting new metrics in each as suggested by [3]. Parameters such as task efficiency and task difficulty
59
in each of the explanatory variables in each game.
• Game types: The Mini-Games can be developed without taking into consideration the mathematical
performance by the implementation of new games to recognize the learning style of each student. For
instance, in [3], it was developed a unique game with different tasks such as collecting and shooting gold
bars in an interactive environment.
• Student mental state: The initial mental state of the student could be taken into consideration by the
use of electrodermal device activity (EDA). The student could spend two minutes in relaxation with
electrodes placed at the middle and ring fingers while listening to calm music [3].
• Different games: The use of games with different game scenarios and avatars according to student
preference. It can be used dynamic difficulty adjustment (DDA) taken into consideration the player’s
emotions such as in [40] with specific threshold values [41]. For instance, this can increase user engage-
ment by the emotional state to discover hidden places in the game, and obtain better results in detecting
the learning style of the student.
• Game experience: Take into consideration the gaming experience related to the time spent in playing
games. The time that each student used per week, month, year; and preference in a game type: RPG,
strategy, simulations, shooters, and serious games [2].
• Demography: Player’s demography can be studied further, including the sex and age of the student
because there exists evidence that boys tend to have more enjoyment in video games than girls [42].
• Computational resources: The computational available resources to test the techniques in the popu-
lation. For instance, the computers have the same specifications such as resolution screen, processor,
RAM, storage, audio, keyboard, and operative system.
• Environment: The environment restrictions in the equipped rooms to test the game, without distractions.
A recommendation was given by a group of researchers who applied the game “Monkey Tales”, that in
three months of game playing, the parents cannot help the student, and children cannot play other video
games in this elapsed time [21].
60
4.3 Conclusions
Learning style recognition is the core to provide personalized education. By the inefficiency in its recognition,
the outcome is poor performance in the students in different subjects. Being one of the most challenging
subjects, mathematics, that is the reason that Latin American countries have one of the lowest scorings in the
international mathematical examination named PISA [7], in comparison with developing countries. By lack of
accessibility to excellent and personalized training in teaching mathematics to children and the motivation that
takes part in it. The use of ICT has been an advantage, but the customized content to each student according to
its learning style is still a challenge. However, finding the best method to learning style detection to the student
it can be a useful tool to diminish this problem.
By the comparison between the four methods: CHAEA-JQ, Mini-Games, ANN, and BN. It was found
that Mini-Games provides percentages close to the most effective traditional one (CHAEA-JQ) to recognize
the most dominant learning style, with an average error ∈= 15.98%, followed by ANN ∈= 18.97%, and
BN ∈= 45.32%. In the same sense, the four techniques show that the Reflector style is the most accurate
in learning style detection, describe as the type of learner that adapts to the traditional educational system.
The error in each technique in this work is by the mechanics in the game mostly, and sample size due by
time constraints. Even other factors such as student mental state, game experience, demography, computational
resources, and environment should be considered to improve the results in this research. Also, dynamic tutorials
in the game, animated avatars in each question in the CHAEA-JQ, and levels in the mathematical Mini-Games
can be improved.
In the predictive technique, the ANN input can be modified by new parameters such as the emotional state
of the student where the target acts as the learning styles percentages using a conventional questionnaire and the
adoption of a new architecture. In the BN, its architecture can be changed and optimize the algorithm to obtain
more validation in the data set. Another fact to take into account is the sample size, due to time constraint in the
project execution. The learning style theory used in this research is based on Kolb’s work. But, further analysis
can provide using the learning style models in another learning theory such as Felderman and Silverman. The
Mini-Games in the learning style detection could be designed to new environments where the student is eager
to play.
As an overall conclusion, the method that recognizes learning styles in comparison with the CHAEA-Junior
questionnaire was the Mini-Games, ANN, and BN in that order. As future work, it is necessary to validate the
experimental design before the testing to provide an accurate probabilistic sampling of the students to support
the results. Besides, the learning styles of detecting models can provide information to provide personalized
content according to the skills of the students by using Mini-Games to increase the mathematical skills in
61
Ecuadorean students.
62
References
[1] A. Y. Kolb and D. A. Kolb, “Lsitechmanual,” pp. 1–72, 2005. [Online]. Available:
papers2://publication/uuid/986F61B1-4EB2-4B91-BBDB-DF326F7EC520
[2] B. Bontchev, D. Vassileva, A. Aleksieva-Petrova, and M. Petrov, “Playing styles based on experiential
learning theory,” Computers in Human Behavior, vol. 85, pp. 319–328, 2018. [Online]. Available:
https://doi.org/10.1016/j.chb.2018.04.009
[3] “Playing style recognition through an adaptive video game,” Computers in Human Behavior, vol. 82, pp.
136–147, 2018. [Online]. Available: https://doi.org/10.1016/j.chb.2017.12.040
[4] S. C. *, “Learning styles: An overview of theories, models, and measures,” Educational Psychology,
vol. 24, no. 4, pp. 419–444, 2004. [Online]. Available: https://doi.org/10.1080/0144341042000228834
[5] O. Chapman, “Challenges in mathematics teacher education,” Journal of Mathematics Teacher Education,
vol. 15, no. 4, pp. 263–270, Aug 2012. [Online]. Available: https://doi.org/10.1007/s10857-012-9223-2
[6] M. Hurst and S. Cordes, Understanding Emotions in Mathematical Thinking and Learning, 1st ed. Aca-
demic Press, 2017.
[7] A. Rivas, Latin America after PISA: Lessons Learned about Education in Seven Countries. CIPPEC-
Natura-Instituto Natura, 2015.
[8] G. D. Vita, “Learning styles, culture and inclusive instruction in the multicultural classroom: A business
and management perspective,” Innovations in Education and Teaching International, vol. 38, no. 2, pp.
165–174, 2001.
[9] N. Zaric, R. Ropke, and U. Schroeder, “A learners’ style under consideration – evaluation and redesign of
a blended learning course in higher education,” 10 2018.
[10] J. Feldman, A. Monteserin, and A. Amandi, “Automatic detection of learning styles: state of the art,”
Artificial Intelligence Review, vol. 44, no. 2, pp. 157–186, 2015.
[11] H. E. Otto and F. Mandorli, “Integration of negative knowledge into mcad education to support
competency development for product design,” Computer-Aided Design and Applications, vol. 14, no. 3,
pp. 269–283, 2017. [Online]. Available: https://doi.org/10.1080/16864360.2016.1240448
63
[12] ——, “A framework for negative knowledge to support hybrid geometric modeling education for product
engineering,” Journal of Computational Design and Engineering, vol. 5, no. 1, pp. 80 – 93, 2018.
[Online]. Available: http://www.sciencedirect.com/science/article/pii/S2288430017300313
[13] G. Baronio, B. Motyl, and D. Paderno, “Technical drawing learning tool-level 2: An
interactive self-learning tool for teaching manufacturing dimensioning,” Computer Applica-
tions in Engineering Education, vol. 24, no. 4, pp. 519–528, 2016. [Online]. Available:
https://onlinelibrary.wiley.com/doi/abs/10.1002/cae.21728
[14] V. Villa, B. Motyl, D. Paderno, and G. Baronio, “Tdeg based framework and tools for
innovation in teaching technical drawing: The example of lamoo project,” Computer Ap-
plications in Engineering Education, vol. 26, no. 5, pp. 1293–1305. [Online]. Available:
https://onlinelibrary.wiley.com/doi/abs/10.1002/cae.22022
[15] A. Latham, K. Crockett, and D. Mclean, “Profiling student learning styles with multilayer perceptron
neural networks,” 2013, pp. 2510–2515.
[16] M. Hasibuan and L. Nugroho, “Detecting learning style using hybrid model,” 2017, pp. 107–111.
[17] M. D. H. E. E. K. Coffield, F., Learning styles and pedagogy in post -16 learning. Learning and Skills
Research Centre, 2004.
[18] N. Stash and P. De Bra, “Incorporating cognitive styles in aha!(the adaptive hypermedia architecture),”
2005.
[19] M. Dziedzic, F. B. De Oliveira, P. R. Janissek, and R. M. Dziedzic, “Comparing learning styles question-
naires,” Proceedings - Frontiers in Education Conference, FIE, no. October, pp. 973–978, 2013.
[20] J. F. S. Delgado, “Chaea-junior survey or how to diagnose elementary and secondary students’ learning
styles,” Journal of Learning Styles, vol. 7, 2013.
[21] E. Nunez Castellar, A. All, L. De Marez, and J. Van Looy, “Cognitive abilities, digital games and arith-
metic performance enhancement: A study comparing the effects of a math game and paper exercises,”
Computers and Education, vol. 85, no. January, pp. 123–133, 2015.
[22] Y. Tazouti, S. Boulaknadel, and Y. Fakhri, “Enhancing young children’s mathematics skills using serious
game approach,” 2015 5th International Conference on Information and Communication Technology and
Accessibility, ICTA 2015, pp. 6–8, 2016.
64
[23] C. C. Chao, “An investigation of learning style differences and attitudes toward digital game-based learn-
ing among mobile users,” Proceedings - Fourth IEEE International Workshop on Wireless, Mobile and
Ubiquitous Technology in Education, WMUTE 2006, pp. 29–31, 2006.
[24] C. Wongwatkit, S. Chookaew, K. Chaturarat, and C. Krutthakha, “An Interactive Story-Based Mobile
Game Application to Diagnosing Learning Style and Learning Suggestion Based on ILS: Development
and Evaluation of the Adventure Hero Game,” Proceedings - 2017 6th IIAI International Congress on
Advanced Applied Informatics, IIAI-AAI 2017, pp. 785–790, 2017.
[25] J. Feldman, A. Monteserin, and A. Amandi, “Detecting students’ perception style by us-
ing games,” Computers and Education, vol. 71, pp. 14–22, 2014. [Online]. Available:
http://dx.doi.org/10.1016/j.compedu.2013.09.007
[26] I. Khan and A. Kulkarni, “Knowledge extraction from survey data using neural networks,” Procedia
Computer Science, vol. 20, 12 2013.
[27] J. J. Lo and P. C. Shu, “Identification of learning styles online by observing learners’ browsing behaviour
through a neural network,” British Journal of Educational Technology, vol. 36, no. 1, pp. 43–55, 2005.
[28] M. Dziedzic, F. Borges de Oliveira, P. Janissek, and R. Dziedzic, “Comparing learning styles question-
naires,” 10 2013, pp. 973–978.
[29] P. Garcıa, A. Amandi, S. Schiaffino, and M. Campo, “Evaluating Bayesian networks’ precision for detect-
ing students’ learning styles,” Computers and Education, vol. 49, no. 3, pp. 794–808, 2007.
[30] L. G. Lopez-Faican and L. A. Chamba-Eras, “Redes bayesianas para predecir el estilo de aprendizaje de
estudiantes en entornos virtuales,” AtoZ: novas praticas em informacao e conhecimento, vol. 3, no. 2, p.
107, 2015.
[31] C. Carmona Marquez, G. Castillo, and E. Millan, “Modelo bayesiano del alumno basado en el estilo de
aprendizaje y las preferencias.” IEEE-RITA, vol. 4, pp. 139–146, 01 2009.
[32] D. A. Georgiou and D. Makry, “A learner’s style and profile recognition via Fuzzy Cognitive Map,”
Proceedings - IEEE International Conference on Advanced Learning Technologies, ICALT 2004, no.
June, pp. 36–40, 2004.
[33] S. V. Kolekar, S. G. Sanjeevi, and D. S. Bormane, “Learning style recognition using Artificial Neural
Network for adaptive user interface in e-learning,” 2010 IEEE International Conference on Computational
Intelligence and Computing Research, ICCIC 2010, no. December, pp. 245–249, 2010.
65
[34] N. Blanchard, M. Brady, A. M. Olney, M. Glaus, X. Sun, M. Nystrand, B. Samei, S. Kelly, S. D. Mello,
N. Dame, and N. Dame, “Artificial Intelligence in Education,” vol. 9112, pp. 23–33, 2015. [Online].
Available: http://link.springer.com/10.1007/978-3-319-19773-9
[35] S. Botsios, D. A. Georgiou, and N. F. Safouris, “Learning Style Estimation Using Bayesian Networks,”
vol. 2007, no. Murray 1999, pp. 415–418, 2011.
[36] B. Bontchev and O. Georgieva, “Playing style recognition through an adaptive video game,”
Computers in Human Behavior, vol. 82, pp. 136 – 147, 2018. [Online]. Available:
http://www.sciencedirect.com/science/article/pii/S0747563217307264
[37] B. Hmedna, A. El Mezouary, O. Baz, and D. Mammass, “A machine learning approach to identify and
track learning styles in moocs,” in 2016 5th International Conference on Multimedia Computing and
Systems (ICMCS), Sep. 2016, pp. 212–216.
[38] R. Torres. (2019) Algorithms implementation. [Online]. Available:
https://github.com/tesisalgoritmo2019/Algoritmos
[39] N. O’Rourke and L. Hatcher, A Step-By-Step Approach to Using SAS System for Factor Analysis and
Structural Equation Modeling, 02 2013.
[40] B. Bontchev, “Adaptation in affective video games,” Cybern. Inf. Technol., vol. 16, no. 3, pp. 3–34, Sep.
2016. [Online]. Available: https://doi.org/10.1515/cait-2016-0032
[41] ——, “Holistic player modeling for controling adaptation in video games,” 04 2016.
[42] A. A. Carvalho and I. Araujo, What Do Portuguese Students Play on Mobile Devices: Inputs for
the Development of Educational Games. Cham: Springer International Publishing, 2016, pp. 69–95.
[Online]. Available: https://doi.org/10.1007/978-3-319-22900-35
66
Appendices
A Appendix 1.
• Tengo fama de decir lo que pienso claramente y sin darle vueltas al asunto.
• Estoy seguro de lo que es bueno y lo que es malo, lo que esta bien y lo que esta mal.
• Muchas veces hago las cosas sin pensar en las consecuencias.
• Me interesa saber como piensan los demas y por que motivos actuan.
• Valoro mucho cuando me hacen un regalo que sea de gran utilidad.
• Procuro enterarme de lo que ocurre en donde estoy.
• Disfruto si tengo tiempo para preparar mis trabajos y hacerlo lo mejor posible.
• Siempre me gusta seguir un orden, en las comidas, en los estudios y hacer ejercicio fısico.
• Prefiero las ideas originales y novedosas aunque no sean muy practicas.
• Acepto y me apego a las normas solo si sirven para lograr lo que me gusta.
• Me gusta mas escuchar que hablar.
• Casi siempre tengo mis cosas ordenadas, porque me disgusta el desorden.
• Antes de hacer algo estudio con cuidado sus ventajas e inconvenientes.
• En las actividades escolares pongo mas interes cuando hago algo nuevo y diferente.
• En una discusion me gusta decir claramente lo que pienso.
• Si juego, dejo los sentimientos por mis amigos a un lado, pues en el juego es importante ganar.
• Me siento a gusto con las personas divertidas aunque a veces me den problemas.
• Expreso abiertamente como me siento.
• En las reuniones y fiestas suelo ser el mas divertido.
• Me gusta analizar las cosas para lograr su solucion.
67
• Prefiero las ideas que sirven para algo y que se puedan realizar, a sonar o fantasear.
• Tengo cuidado y pienso las cosas antes de sacar conclusiones.
• Intento hacer las cosas para que me queden perfectas.
• Prefiero oır las opiniones de los demas antes de exponer la mıa.
• En las discusiones me gusta observar como actuan los demas participantes.
• Me disgusta estar con personas calladas y que piensan mucho todas las cosas.
• Me angustia si me obligan a acelerar mucho el trabajo para cumplir un plazo.
• Doy ideas nuevas y espontaneas en los trabajos en grupo.
• La mayorıa de las veces creo que es preciso saltarse las normas, mas que cumplirlas.
• Cuando estoy con mis amigos hablo mas que escucho.
• Creo que, siempre, deben hacerse las cosas con logica y de forma razonada.
• Me ponen nervioso/a aquellos que dicen cosas poco importantes o sin sentido.
• Me gusta comprobar que las cosas funcionan realmente.
• Rechazo las ideas originales y espontaneas si veo que no sirven para algo practico.
• Con frecuencia pienso en las consecuencias de mis actos para prever el futuro.
• En muchas ocasiones, si deseo algo, no importa lo que se haga para conseguirlo.
• Me molestan los companeros y personas que hacen las cosas a lo loco.
• Suelo reflexionar sobre los asuntos y problemas.
• Con frecuencia soy una de las personas que mas animan las fiestas.
• Los que me conocen suelen pensar que soy poco sensible a sus sentimientos.
• Me cuesta mucho planificar mis tareas y estudiar con tiempo para mis examenes.
• Cuando trabajo en equipo me interesa saber lo que opinan los demas.
• Me molesta que la gente no se tome las cosas en serio.
• A menudo me doy cuenta de otras formas mejores de hacer las cosas.
68