Evolutionary Algorithm Based Rule(s) Generation for

Personalized Courseware Construction in Educational

Data Mining
Sreenath. K1, Jeyakumar. G2
Department of Computer Science and Engineering,
Amrita School of Engineering, Coimbatore
Amrita Vishwa Vidyapeetham, Amrita University, India
1
[email protected], [email protected]

Abstract—Evolutionary Computing for Educational Data than the classical greedy techniques to arrive at stronger
Mining is a research field which with the applications of solutions [2]. The use of evolutionary algorithms for mining
Evolutionary Algorithms (EAs) to mine, analyze and modify data from the educational database is an emerging research
educational data. This paper presents the most relevant studies area. Popular evolutionary algorithms used are Genetic
conducted in this research area. The paper also describes
different EAs used for implementing different data mining
Algorithm (GA), Genetic Programming (GP) and Differential
techniques. It goes on to list how these algorithms are utilized by Evolution (DE) and the common data mining techniques used
different educational users to carry out different tasks. Finally, a are classification, clustering, association rule mining and
new combination of EA, Educational User and data mining summarization.
technique is suggested for implementation. As a part of that a
personalized courseware construction technique is proposed and II. EDUCATIONAL DATA MINING
a sample courseware is constructed using the proposed Educational Data Mining (EDM) is a research area that is
technique. The details about the rule construction and the data concerned with applying different data mining techniques to
mining process involved in the courseware construction
extract data from the educational databases, and using mined
techniques are also explained.
data to better understand and improve the teaching-learning
Keywords—Evolutionary algorithms; Educational Data processes in educational environments. The EDM has
Mining; Courseware Construction; Association Rule Mining; following four key areas of application [3],
Differential Evolution. 1) Improving student models: student models describe a
student’s characteristics such as the student’s
I. INTRODUCTION attitudes, current knowledge and motivation.
2) Improving domain models that involve developing
Evolutionary Algorithms are the optimization tools that are algorithms for determining the knowledge structure
based on Darwinian principles of evolution. An EA works models which explain the knowledge
with the biological evolution mechanisms like reproduction, interrelationships in a domain.
mutation, crossover, selection. Initially, a population of 3) Evaluating the academic support provided by the
candidate solutions is formed randomly then new and better learning tool, which involves course and teaching
solutions are evolved from these candidates using the effectiveness; and
biological processes. The major motivation for using 4) Scientific research into learning and learners to
evolutionary algorithms for data mining tasks is that they obtain deeper understanding of the key aspects
utilize powerful and efficient search methodologies in order to impacting learning.
carry out a global search in the solution space. These The EDM can be defined by describing different groups of
algorithms, by various research works, have been found to users and the tasks that the user is required to perform. The
cope better than the classical greedy data mining techniques to different users involved are:
arrive at solutions. 1) Learners / Students,
Educational Data Mining (EDM) is a data mining research 2) Educators,
area concerned with creating methods for exploring different 3) Course authors,
classes of data from various educational information 4) Organizations and
structures, and using them to better understand students, 5) Administrators.
teacher and the teaching-learning process[1]. The main There are a number of tasks that an educational environment
purpose of educational data mining is to convert the raw resolves through different data mining techniques. Some of the
educational data into useful information to analyze and tasks performed by different users are:
interpret different educational research issues. Since data 1) Providing feedback for teachers: The main aim of
mining processes are usually approximation problems, it has providing feedback is to improve the instructors'
been found that the evolutionary algorithms tend to cope better decision making and to make appropriate actions.

978-1-5090-0612-0/16/$31.00 ©2016 IEEE

 2016 IEEE International Conference on Computational Intelligence and Computing Research

Online based quiz systems are becoming extremely survey discusses strategies used by the authors to use EAs for
popular for obtaining quick and honest feedback from EDM.
the students. Recently use of multiple choice Hypermedia Educational Systems are online systems that
questions for feedback has also become popular enable the student to have a course based on his/her interests
where students are asked to choose the best possible and qualities. A method to alter the contents of the course is
reaction [11]. Using these responses, a large amount proposed by Romero [4]. Association rules are created based
of knowledge can be mined about the students. The on the results taken from the evaluation of a Hypermedia
most commonly used data mining methods for this Course on Rheumatology. Using GA, every individual encodes
purpose are association rule mining, clustering, a solution and evolves into a more effective individual by
classification and pattern analysis. reproduction. Using an appropriate selection method, here
2) Course recommendations for students: Elitist selection, strongest individuals are retained and the
Recommending relevant and useful courses for the weak individuals are selected for reproduction. Mutation
students is a very crucial task for an e-learning involves creating a new rule from an existing rule by changing
system. The system has to recommend appropriate a value or variable. Crossover involves making new
courses for the students based on their search history individuals from existing rules by interchanging the
and previous courses so that new and similar courses consequent and antecedent. The rules which have the highest
in the system are taken up by the students. support and confidence values are retained through each
Association rule mining and clustering are commonly generation. The paper goes on to suggest that GA is an
used for this purpose. efficient technique for extracting a small set of comprehensive
3) Grouping students: The e-learning systems attract a rules.
large number of students from all over the world. It GA has been found to be one of the major influences in the
becomes a necessity to group students based on their data mining area. Minaei-Bidgoli uses GA [5] for
performance, characteristics and interests. To group classification technique for classifying students on the basis of
the students, based on their features, classification their marks. Classifying students is one of the most widely
and clustering are used. used features of educational data mining. Initially the training
4) Predicting student performance: The main set and the classes are found from the data. A feature vector is
functionality of predicting student performance is to an individual which is formed by the values of six variables
calculate a value which is to illustrate the features of defined based on the time, rate of success and failure and
a student. The data mining techniques like regression, student interactions. Each individual is formed based on these
classification and clustering predict student features with each taking either 0 or 1. The population is
performance based on the quantitative information initially formed randomly and reproduction process is
provided. performed on the population. The fitness function is the error
5) Student modelling: Student model ling is the process rate achieved between the training data and each individual.
of developing models for students based on their skill The individuals with highest fitness are preserved for the next
and knowledge. Student models assess a range of generation and the others are selected for reproduction. After a
features of a student such as behavior and number of iterations, the new testing data is given to the
motivational character. Clustering and classification system and the performance of new student is predicted.
are the major data mining techniques used for this Subgroup discovery for predicting student performance is
purpose. one of the widely used tasks by an institution. The research
6) Courseware modelling and maintenance: The main works reported in [6, 7] suggest algorithms to classify students
objective of this task is to help the instructors and the based on association rules formed from the data. These
authors in the development and maintenance of their techniques are extensively used for educational purposes for
courseware. It also attempts to reuse the existing mining data to better understand learning techniques and get
resources to different users and learners. Association inferences about the students. Rules which describe subsets of
rules and clustering have been widely used by the the population are formed to discover subgroups. These rules
developers in order to develop different arguments describe the relationships between the students’ utilization of
for construction and maintenance of courseware. the various modules and activities delivered by this e-learning
system and the grades procured in the courses. GA is used for
III. LITERATURE SURVEY inducing the fuzzy rules formed from the data. Fitness
Research works in Evolutionary Algorithms for educational measure is calculated using accuracy, coverage and
data mining have been taking place since the inception of significance values. It compares the experimental results of
EDM. Given most institutions across the globe use some form subgroup discovery for Moodle e-learning data in the classical
of educational data mining, it is very crucial that the data algorithms and EAs. Apriori-SD and CN2-SD are the two
mining methods be updated so as to attain the maximum classical data mining algorithms used for subgroup discovery
efficiency. A number of research works have found that the and SDIGA and MESDIF are the two evolutionary algorithms
usage of evolutionary algorithms have contributed greatly to used. In the experiment, 5 courses have been chosen involving
the effectiveness of different data mining techniques. This a total of 293 students. The rules are obtained for all the
 2016 IEEE International Conference on Computational Intelligence and Computing Research

algorithms. SDIGA and MESDIF employ fuzzy rules to classical data mining algorithms to GP and found that GP
discover the subgroups. Then the experiment is conducted and discovers more interesting rules than classic algorithms. One
the quality measures are recorded. Comparing the results of of the advantages the GP has over the classical algorithms is
the experiment, suitability of EA to this problem was found. It the ability of flexible capacity to represent the individuals.
was also found that the EA offer information with a lesser Another application is in the use of Multiple-choice Quiz
number of variables and rules. (MCQ) for knowledge discovery. MCQ is one of the most
To predict student performance, classification technique widely used forms of tests in education [11]. The data
which integrates GA and Support Vector Machine (SVM) obtained from MCQs provide numerous information such as
algorithm was introduced by Malviya [8]. Sequential Minimal scores for each question and the time taken for each question
Optimization (SMO) algorithm is the training algorithm in etc. Association rule mining is a widely accepted method for
WEKA tool for SVM classification. The SMO algorithm divides mining the educational data. This paper introduces a new
the solution of a large quadratic programming optimization technique of combining grammar-based GP and association
problem into a number of smaller QP problems. Using the rule mining to extract information from the database. During
SMO algorithm, set with selected feature subset and value of pre-process, matrices are generated from the data. A new
parameters are trained. The population based on the fitness
method called Grammar-Guided GP Association Rule Mining
values computed from the SVM classifier is selected. By
(G3PARM) is used in this paper to mine the data. This has
applying the genetic operations the process in continued until
the optimal feature subset is obtained. The new method been found better than the classical association rule mining
(SMO+GA) described was found to remove the insignificant algorithm as it produced lesser number of rules but with
features and effectively find best parameter values. The new highest support and confidence. The course feedback can be
method was compared with the classical SMO algorithm, and obtained to assess the effect of applying the updates in the
was found that the correctly classified items are increased in quiz. The course can be updated with the new rules which
the new algorithm. have the best confidence-support ratio.
Predicting student failure [12] at institutions has become
Decision tree is a decision support tool that describes rules one of the most challenging tasks for the institution since it is
through trees. A combination of decision trees and GA is used affected by a number of factors. So pre-processing becomes
to predict the student course performance [9]. Decision tree is one of the most important tasks to be carried out during
a very popular method used for classification. Using decision mining process since it decides attributes to be selected for
trees, the information contained in the data is broken down prediction. WEKA tool [13] provides feature selection
and sub divided to discover rules and relationships. Each tree algorithms to perform attribute selection. The decision trees
is considered as a chromosome in a population. Applying the are then formed and the GP algorithm is applied to the
genetic operators, new generations are consequently obtained population. The algorithm is compared with the classical data
and the individuals with best fitness values are found. The mining approaches. The true positive rate, true negative rate,
total number of misclassification is taken as the fitness accuracy of the evolutionary data mining was found to be
function. The simple GA (SGA) is performed on the data and more than the classical algorithms.
the efficiency is calculated. It is then compared with the Building a student performance prediction model is one of
normal genetic algorithm to assess the efficiency. The the most widely used task in educational data mining. In [14],
Association-based GA (AGA) achieves greater prediction for this purpose variables based on Subject, Division of Labor,
accuracy in lesser computation time for classification than Rules are constructed. Based on these variables decision trees
simple GA when evaluated using five-fold cross validation. are made to form the individuals in GP. The genetic operators
This study helps the teachers to improve the student are applied to the candidates. A confusion matrix of the model
categorizing, and also determine the students who require is calculated to allow comprehensive analysis of the prediction
individual attention. performance. Sensitivity and specificity are used as measures
Genetic Programming (GP) is another evolutionary of fitness obtained from the matrix. RandomTree algorithm is
algorithm which is commonly used to implement data mining used to create the tree structures for decision trees. A
methods. Unlike GA, GP makes use of tree structures to form hierarchical structure is created based on the conditions which
each individual. Knowledge discovery using GP for providing optimizes the algorithm to predict student performance.
feedback to courseware authors is another application of Another work where student performance prediction is the
evolutionary algorithms [10]. The main aim of this work is to objective is given by Zafra and Ventura [15]. It uses Multiple
recognize powerful dependence relationships within the data Instance Learning (MIL) which generates a classifier that
gathered from various students’ sessions. The data initially is classifies unseen patterns. An extension of MIL combined with
pre-processed for mining, using the tasks: attribute selection, GP (G3P-MI) was introduced. The attributes are initially
data cleaning, transforming continuous attributes and data selected for the system and the rules are formed by the
integration. This pre-processing is done to extract a subset of attributes. Methods based on supervised learning, logistic
valid rules from the whole set of rules discovered. Association regression, neighbour distance are assessed with GP.
rules are formed from the data and the trees are created to Accuracy, sensitivity and specificity are used to compare the
form the individuals of the population. Then the algorithm is different techniques and results were found in favour of EA. A
applied to the population. This paper goes on to compare the comparison is made between the classical multiple instance
 2016 IEEE International Conference on Computational Intelligence and Computing Research

learning algorithms and G3P-MI and better accuracy, is to assist the teachers and course authors to realize the
sensitivity and specificity were found for G3P-MI. A multi- process of developing courseware and automatic content
gene genetic programming is used for predicting student learning. Based on the student feedbacks, evaluations, the
failure rate [16]. A multi-gene individual is a tree consisting of methods the student use to learn, the course needs to be
multiple attributes which form each individual in the upgraded constantly.
population. Here the number of genes is four. The population According to recent surveys, the research line of
is formed and the genetic operators are applied. construction and maintaining courseware has had the
Student modelling is another direction of research in minimum number of papers published [17]. Hence we choose
Educational Data Mining. The authors in [22] generated the the task for our problem as: Courseware construction and
learning curve for showing the variations in students’ learning maintenance. A figure comparing number of research works in
capacity in the class room. An attempt to predict the learning each of EDM task is shown in Figure 1. As well as, most of
style of individual students in a class through on-line games is the research works which implement this task use the classical
done in [23]. As an approach to model human brain, a data mining algorithms. García [18, 19] implemented methods
simulated computational model for human brain is presented of Apriori algorithm in order to form different association
in [24]. rules for knowledge discovery. These rules are used to
recommend courseware authors on the changes to make in the
IV. SUMMARY OF SURVEY current system.
The authors have reviewed the research works which used EAs Experiments have proved that most evolutionary
for different EDM purposes. The uses EAs, through various algorithms tend to extract better and relevant data than the
experiments, have been proved to be better than the classical classical data mining techniques. So, it is evident from the
data mining techniques. Also, hybridization of EAs and the literature that EAs are used for accurate knowledge extraction.
data mining techniques provide far better results. The details GA and GP are found to be widely used for educational data
of the most relevant research works in this field are detailed in mining tasks.
this survey. Many research works in solving optimization problems
One of the major decisions to take when designing an have arrived at the conclusion that DE were found to be better
EDM system is to choose the task the system performs. There than GA and GP [20]. Studies show that that DE has a greater
are a number of tasks that an EDM system can perform. degree of advantages than GA [25, 26]. Hegerty [20] validated
Romero [17] defines a number of categories of tasks that can this for solving N-queens problem and travelling salesman
be done by an EDM system. Main categories described are problem. The experiments showed that the optimum solutions
listed in Section 2. Apart from them the two other important were found in lesser generations by DE than GA. Vesterstrom
tasks are: Recommendations for authors and providing [21] compared DE with Particle Swarm Optimization (PSO)
feedback for authors. A problem in EDM is categorized by algorithm with 34 benchmark problems. It was found that the
four factors: mean and standard deviations for DE were found to be better
1) The task: providing feedback, course than PSO and other EAs. Vesterstrom goes on to state that DE
recommendations, grouping students, predicting is a simple, fast and robust and finds the optimum in almost
student performance, student modelling and every run. Moreover, it has only a very less number of
courseware maintenance. parameters to be set, and the same parameters can be utilized
2) The user: Students, Educators, Course authors, for a number of different problems. Based on these insights
Organizations and Administrators. about DE, we chose DE for our experiment.
3) The data mining technique: classification, clustering, The courseware authors have to regularly update the
association rule mining courseware for improving the performance of the system. The
4) The algorithm: Apriori Algorithms, EAs, etc recommendations for the authors are an important aspect for
The Table 1 summarizes the factors involved in each of the an educational system. Hence, we choose the course authors
paper. The papers discussed in section 3 also have proved that as user for our problem. Finally, since association rules are
using EAs for EDM tasks is robust and efficient in terms of simple and efficient tool for analyzing and predicting student
providing required mining with less number of features and behaviour, we selected association rules as our data mining
improved accuracy. technique.
The objective of this survey on EAs for EDM is to identify Thus, based on the summary of literature survey this paper
a problem to solve a complicated EDM task by an efficient proposes to implement a system for “courseware construction
EA. This also has an intension to prove the robustness and for course authors, using association rule mining and
simplicity of the EAs in hard EDM task. Identification of a differential evolution algorithm”. As a preliminary step
problem in EDM involves the identification of the above said towards this direction of work a sample courseware
four factors (task, user, technique and algorithm). construction using the tool KEEL is done and the observations
Among all the EDM tasks the courseware (course material) are reported in next section.
related tasks need special attention. The objective of
constructing/developing courseware, as given by Romero [17],
 2016 IEEE International Conference on Computational Intelligence and Computing Research

TABLE I. THE SUMMARY OF FACTORS IN EDM

Sno Ref. No. Task User Data Mining Technique Algorithm

Improve performance Teacher Association rules Genetic Algorithm
1 [4]
Predicting Student Performance Institution Classification Genetic Algorithm
2 [5]
Grouping Students Management Classification using rules. Genetic Algorithm
3 [6] (for teachers)
Predict final grade Institutions Classification Genetic Algorithm
4 [7]
Students performance Institutions Classification(SVM) Genetic Algorithm
5 [8]
Predicting the student performance Institution Association Rules Genetic algorithm
6 [9]
Providing feedback to improve Courseware authors Association Rules Genetic Programming
7 [10] courseware
Providing Feedback Instructors Association Rule Mining Genetic Programming
8 [11]
Detecting student failure Institutions Classification using association rules Genetic Programming
9 [12]
Student performance prediction to Instructors Decision trees Genetic Programming
10 [14] monitor students’ progress
Predict a student's performance Institutions Prediction rules Genetic Programming
11 [15]
Predict student failure rate to Educational Decision trees Genetic Programming
12 [16] evaluate the performance of curricula administrators

Fig. 1. The Number of Paper in Each Task (Figure courtesy, [4])

A courseware construction technique where students are

V. SAMPLE COURSEWARE CONSTRUCTION divided into different clusters based on students' academic
This section initially presents a survey of various existing history is proposed in [28]. A web tutor tree is formed using
courseware construction techniques in the literature and this data, and algorithms are suggested to form personalized
discusses the construction of a sample courseware based on trees. Each node in the tree has information about the students,
the problem statement. represented in weights. A concept hierarchy technique is
described in [29] which divide the contents into concepts. The
 2016 IEEE International Conference on Computational Intelligence and Computing Research

online processing step is described in [30], is an adaptive This rule shows that if the intuitive learning style of a
system for authors based on their needs. A profile of the user student is within the range of the score mentioned then he/she
is used in order to recommend contents for the students in [29] tends to be a sensitive learner. Hence, if conceptual materials
and [30]. The use of clustering methods to categorize are given as a courseware, then he/she can be given
courseware materials are proposed in [31]. It is a tool based on questionnaires and presentations also as materials.
web Semantics framework. It is implemented with the aim of
assisting users to organize courseware resources with the IF "verbal" value="[3.333, 3.543]" AND "sens"
usage of Bisection K-Means algorithm. Clustering is used to value="[2.142 2.875]"
group learning documents according on their topics and THEN "intui" value="[2.857, 3.529]"
similarities. A graph system called Document Index Graph
(DIG) is introduced for representation of documents. In [32], This rule indicates that if the verbal and sensitive learning
the author discusses a method where the materials suitable for styles are within the respective score range then he/she is
different learning styles of students are framed. marked for intuitive learning style. Thus, adding discussions
A system based on Felder-Silverman learning style is and questionnaires would augment the concepts included in
introduced in [23] to determine the learning style of individual the courseware. The suggestions to construct the courseware
students in a class. Our experiment uses the database are derived from the association rules formed by the KEEL
constructed in [23] as a knowledge source to mine the learning tool. Accordingly sample coursewares are constructed and
styles of students and/or classes at different levels. verified the correctness of the content of the courseware with
From the individual learning styles, we focus on to mine individual students.
the learning style of the whole class as well as the learning In courseware construction of a subject a pool of suitable
style of meaningful groups in a class such as Top Performers, text documents (.pdf, .docx and .txt etc), audio files, video
Average Performers and Below Average Performers. We use files, images, presentations and questionnaires are to be made
association rule mining to identify learning styles at different available in a source. Now, based on the rules generated and
levels. The association rules generated by the proper EA are the learning style identified by the system different
used as the basis for constructing the courseware. courseware can be suggested for the different students.
This work initially focuses on constructing personalized
courseware for individual students using association rule VI. CONCLUSIONS
mining. The learning styles defined by the system [23] are This paper has presented the details of the papers reported in
visual, verbal, active, reflective, sequential, global, sensing the literature for using Evolutionary Algorithms for
and reflective. The system developed in the document Educational Data Mining. Even though there was less number
evaluates the learning style of each of the student through of reports found in this field of research, the authors have
different games and assign scores accordingly. Each student in presented all the papers in detail. The summary of four factors
a class obtains a score for every learning style based on the (task, user, technique and algorithm) with respect to all the
level they complete in each of the corresponding game. papers presented in this paper would be useful to the readers
Now using these scores, we propose to form strong interested in doing research in this direction. As an initial
association rules to produce a new measure for their learning attempt in this direction of research, a personalized
style. The rules generated by the proposed idea would give courseware construction strategy is proposed. The proposed
insight about strengths and weaknesses of a student’s learning strategy is implemented and tested using the KEEL tool. Set of
style. Using these parameters, appropriate courseware association rules are generated from the student learning style
(learning materials) can be constructed and provided to the database, suggestions to construct the courseware are derived
students. from the rules and sample courseware are constructed for
KEEL is a software tool to implement different EAs for students based on their learning styles.
different problems in data mining. We used KEEL as tool for
our experiment. Since GA is very native to KEEL tool, we REFERENCES
initially used GA for rule generation. The tool, using GA, [1] On line resource: http://www.educationaldatamining.org/
formed various rules by defining relationships between [2] Freitas and A. Alex, "A review of evolutionary algorithms for data
different learning styles of the students. The rules are mining." Soft Computing for Knowledge Discovery and Data Mining.
Springer US. pp. 79-111, 2008.
generated based on the support and confidence achieved for
[3] Baker, Ryan SJD, and Kalina Yacef. "The state of educational data
every rule. The average support value is 0.36 and the mining in 2009: A review and future visions." JEDM-Journal of
confidence value is 1.0. Based on the different learning styles, Educational Data Mining1.1 pp. 3-17, (2009).
personalized courseware is constructed. A sample rule [4] C. Romero, S. Ventura, C. De Castro, W. Hall, & M. H. Ng. Using
constructed in our experiment is show below genetic algorithms for data mining in web-based educational hypermedia
systems. In Workshop on Adaptive Systems for Web-based Education
(2002, May).
IF "intui" value=" [2.857, 3.529]" [5] Minaei-Bidgoli, Behrouz, and William F. Punch. "Using genetic
THEN "sens" value="[2.142, 2.875]" algorithms for data mining optimization in an educational web-based
system." Genetic and Evolutionary Computation—GECCO 2003.
Springer Berlin Heidelberg, 2003.
 2016 IEEE International Conference on Computational Intelligence and Computing Research

[6] C. Romero, P. González, S. Ventura, M. J. Del Jesús, & F. Herrera. : Computer Networks and Security – ELSEVIER – Ed. K.R.Venugopal,
Evolutionary algorithms for subgroup discovery in e-learning: A S.C Lingardd,y BANGALORE, India, pp. 152- 160, 2014.
practical application using Moodle data. Expert Systems with [23] G. Jeyakumar, K. Raaghul, N. Pavithra, D. Kavya and B. Aswathi, ”A
Applications, 36(2), pp. 1632-1644 (2009). Prototype for Student Learning Style Modelling Using Felder-Silverman
[7] C. J. Carmona, P. González, M. J. del Jesus, C. Romero, & S. Ventura. Learning Style Model,” In Proceedings of International Conference on
Evolutionary algorithms for subgroup discovery applied to e-learning Smart Structures & Systems (ICSSS-2016), Chennai, India, 23rd March
data. In IEEE EDUCON 2010 Conference (pp. 983-990). IEEE (April – 2016, pp. 178 – 182.
2010). [24] K. Roshini, B. Bavya and G. Jeyakumar, “RoBaJe – A Simulated
[8] Prof. Kamiya Malviya, and Anurag Jain. "A Hybrid Approach for Computational Model for Human Memory to Illustrate Encoding and
Integrating Genetic Algorithms with SVM for Classification and Decoding of Information”, International Journal of Applied Engineering
Modelling Higher Education Data.” International Journal of Science Research, Research India Publications, ISSN 0973-4562 Vol. 9, No. 24,
and Research (2015). pp. 26957-26970, January-2015.
[25] G. Jeyakumar, and C. ShunmugaVelayutham, “A Comparative
[9] Hsu, Pei-Lun, Robert Lai, and C. C. Chiu. "The hybrid of association
Performance Analysis of Differential Evolution and Dynamic
rule algorithms and genetic algorithms for tree induction: an example of
Differential Evolution Variants”, Proceedings - World Congress on
predicting the student course performance." Expert Systems with
Nature and Biologically Inspired Computing, pp. 463-468, 2009.
Applications 25.1 (2003): pp. 51-62.
[26] G. Jeyakumar, and C. ShunmugaVelayutham. “An Empirical
[10] C. Romero, S. Ventura, & P. De Bra. Knowledge discovery with genetic
Comparison of Differential Evolution Variants on Different Classes of
programming for providing feedback to courseware authors. User
Unconstrained Global Optimization Problems”, Proceedings -
Modeling and User-Adapted Interaction, 14(5), pp. 425-464 (2004).
International Conference on Computer Information Systems &
[11] C. Romero, A. Zafra, M. J. Luna, & S. Ventura. Association rule mining
Industrial Management Applications, pp. 866-871, 2009.
using genetic programming to provide feedback to instructors from
[27] Alcalá-Fdez, Jesús, Luciano Sanchez, Salvador Garcia, Maria Jose del
multiple choice quiz data. Expert Systems, 30(2), pp. 162-172 (2013).
Jesus, Sebastian Ventura, Josep Maria Garrell, Jose Otero et al. "KEEL:
[12] C. Márquez-Vera, A. Cano, C. Romero, & S. Ventura. Predicting
a software tool to assess evolutionary algorithms for data mining
student failure at school using genetic programming and different data
problems." Soft Computing 13, no. 3 (2009): pp. 307-318.
mining approaches with high dimensional and imbalanced data. Applied
[28] C. Tang, R. W. Lau, Q. Li, H. Yin, T. Li, & D. Kilis. Personalized
intelligence, 38(3), pp. 315-330 (2013).
courseware construction based on web data mining. In Web Information
[13] Holmes, Geoffrey, Andrew Donkin, and Ian H. Witten. "Weka: A
Systems Engineering, 2000. Proceedings of the First International
machine learning workbench." Intelligent Information Systems, 1994.
Conference on (Vol. 2, pp. 204-211). IEEE, 2000.
Proceedings of the 1994 Second Australian and New Zealand
[29] Cristea, Alexandra, and Lora Aroyo. "Adaptive authoring of adaptive
Conference on. IEEE, 1994.
educational hypermedia." International Conference on Adaptive
[14] W. Xing, R. Guo, E. Petakovic, & S. Goggins. Participation-based
Hypermedia and Adaptive Web-Based Systems. Springer Berlin
student final performance prediction model through interpretable
Heidelberg, 2002.
Genetic Programming: Integrating learning analytics, educational data
[30] Alfonseca, Enrique, Pilar Rodríguez, and Diana Pérez. "An approach for
mining and theory. Computers in Human Behavior, 47, pp. 168-181
automatic generation of adaptive hypermedia in education with
(2015).
multilingual knowledge discovery techniques." Computers &
[15] Zafra, Amelia, and Sebastián Ventura. "Predicting Student Grades in
Education 49.2 (2007): pp. 495-513.
Learning Management Systems with Multiple Instance Genetic
[31] N. Venkatesan. "Role of data mining techniques in educational and e-
Programming." International Working Group on Educational Data
learning system." Asia Pacific Journal of Research (Vol: II, Issue: VIII),
Mining (2009).
2013.
[16] J. O. Orove,, N. E. Osegi, and B. O. Eke. "A Multi-Gene Genetic
[32] A. L. Franzoni, S. Assar, B. Defude, & J. Rojas. Student learning styles
Programming Application for Predicting Students Failure at
adaptation method based on teaching strategies and electronic media.
School." arXiv preprint arXiv: 1503.03211 (2015).
In 2008 Eighth IEEE International Conference on Advanced Learning
[17] Romero, Cristóbal, and Sebastián Ventura. "Educational data mining: a
review of the state of the art." Systems, Man, and Cybernetics, Part C: Technologies (pp. 778-782). IEEE (2008, July).
Applications and Reviews, IEEE Transactions on 40.6 (2010): pp. 601-
618.
[18] E. García, C. Romero, S. Ventura, & C. De Castro. An architecture for
making recommendations to courseware authors using association rule
mining and collaborative filtering. User Modeling and User-Adapted
Interaction, 19(1-2), pp. 99-132 (2009).
[19] E. García, C. Romero, S. Ventura, & C. de Castro. Using rules discovery
for the continuous improvement of e-learning courses. In International
Conference on Intelligent Data Engineering and Automated
Learning (pp. 887-895). Springer Berlin Heidelberg (2006, September).
[20] Hegerty, Brian, Chih-Cheng Hung, and Kristen Kasprak. "A
comparative study on differential evolution and genetic algorithms for
some combinatorial problems." Proceedings of 8th Mexican
International Conference on Artificial Intelligence. 2009.
[21] Vesterstrom, Jakob, and Rene Thomsen. "A comparative study of
differential evolution, particle swarm optimization, and evolutionary
algorithms on numerical benchmark problems." Evolutionary
Computation, 2004. CEC2004. Congress on. Vol. 2. IEEE, 2004.
[22] V. Sucharitha, R. ReshmaReddy, G. Jeyakumar, “Learning Interest
Curve generation using PECS Agent based Model”, International
Conference on Communication and Computing (ICC2014), Proceedings