IEEE Sponsored 2nd International Conference on Innovations in Information Embedded and Communication

Systems ICIIECS’15

Applications of Educational Data Mining: A

Survey
S. Hari Ganesh A. Joy Christy
Asst. Professor, Department of Computer Applications Research Scholar, Department of Computer Science
Bishop Heber College (Autonomous) Bishop Heber College (Autonomous)
Tiruchirappalli, India Tiruchirappalli, India
Email: hariganesh17@gmail.com Email: joychristy001@gmail.com

Abstract: Various educational oriented problems are II. LITERATURE SURVEY

resolved through Educational Data Mining, which is the A.EDM Contributions
most prevalent applications of data mining. One of the Tripti et al [3] focused on the early prediction
crucial goals of this paper is to study the most recent of students who are at the risk of failure thereby helping
works carried out on EDM and analyze their merits and
the teacher to take timely action to improve the
drawbacks. This paper also highlights the cumulative
results of the various data mining practices and students’ performance through extra coaching and
techniques applied in the surveyed articles, and thereby counseling. They also classified the important attributes
suggesting the researchers on the future directions on that influenced students’ third semester performance
EDM. In addition, an experiment was also conducted to and established the effects of emotional quotient
evaluate, certain classification and clustering algorithms parameters that influenced placement.
to observe the most reliable algorithms for future This paper recommended three significant
researches. features and that were found to be valuable in student
Keywords: EDM, Techniques, approaches, Data Mining. performance prediction
I. INTRODUCTION • Second semester result is a key predictor of third
Almost all the data in today’s world are kept in semester
databases. The term database not only refers to the • A good academic track is a good indicator of good
storage of data, but also acts as a source of hidden and performance
unpredictable information that do not go wasted. Data • Emotional attributes affects the students placement
Mining (DM) is a flourishing discipline in computer performance
society that mines useful and interesting data from large This work could have been tested with B.E,
data repositories, generates meaningful facts and B. Tech students who were often likely to be dropped
discerns knowledge. Data mining techniques and out from the college. This system was not developed
concepts have been applied in many fields including with the decision support system to help authorities
health, finance, commercial, scientific etc. identify weak students and take timely measures.
A new application of DM called Educational Cristobel Romero et al [4] investigated how
Data Mining (EDM) encompasses in extracting and different data mining approaches can be used to
interpreting data that comes from educational improve the prediction of first-year computer science
background or domain. EDM incorporates with an university students’ final performance on the basis of
examination of many educational facts that is fruitful their participation in an on-line discussion forum. They
for the educators to afford students qualitative introduced a new approach of classification via
education. The EDM works by converting the raw data clustering instead of traditional classification
coming from educational systems into useful algorithms to predict two classes of pass or fail.
information that could potentially have a great impact The result showed, that the use of data from
on educational research and practice [1]. In general the middle of the course allows for early prediction that
EDM designs models, techniques, tasks, and algorithms alerts tutors about those students who are potentially at
to explore educational data [2]. the risk of failing. One disadvantage that the tutors
The rest of this paper is organized as follow: Section 2 experienced in the system was the evaluation of
describes literature survey, Section 3 explains the students’ messages, as it was so tedious and time
common EDM practices finally Section 4 elucidates the consuming. They would have been suggested an
conclusion this work. automation system for the above said problem.

978-1-4799-6818-3/15/$31.00 © 2015 IEEE

 IEEE Sponsored 2nd International Conference on Innovations in Information Embedded and Communication
Systems ICIIECS’15

Saurabh Pal [5] implemented certain data designing a data model for storing the activity data and
mining methodologies to find students who are likely to creating modules to monitor and visualize learner
be dropped out from their first year B.E program. The viewing behavior.
author used Naïve Bayes classification algorithm for The findings showed only a small number of
the accurate prediction of attributes from the existing students viewed videos regularly and in the proposed
data sets. The results exhibited that the machine order. Most students viewed the videos in the last few
learning algorithm compared the new data set with the days before the tests. Moreover, video sections were
existing and was able to establish effective dropout viewed mostly in sequence. Pausing and resuming was
predictive model. mainly observed for videos that were associated with an
The system produced almost accurate assignment. The findings also revealed design problems
prediction in identifying the students who needed related to some interactive items that caused an
special attention to reduce drop-out rate. But the result unexpected number of unsuccessful attempts. The
was too short and tested only with the small dataset. author did not analyze the reason that affects learner
Leila Dadkhahan [6], justified the needs for viewing behavior and the factors that cause refraining
student retention in higher education institutions, from online videos.
presented certain theoretical student retention models, Suhem et al [9], discussed the application of
and also assessed the previous works to show how data data mining in education for student profiling and
mining techniques could be applied to discover new grouping. They applied Apriori algorithm to the
knowledge that increases the retention rate. The study database containing academic records of various
comprehended that the first one year is very critical students and tried to extract association rules in order to
which increases the retention rate as the students who profile students based on various parameters like exam
continue to their sophomore year are more likely to marks, result grades, attendance and practical exams.
graduate. They also applied K-means clustering to the same data
The extracted knowledge helped institutions in set to group the students. The conclusion stated the use
building proper intervention programme and they of data mining techniques easily clustered the students,
applied it at the right time to those students in need of identified hidden patterns about their learning styles,
these programme. As a result, students’ academic found undesirable student behavior and performed
performance improved and leaded to increase student student profiling.
retention and graduation rate for the institutions. This The results enormously deduced the manual
work only compared the existing algorithms, works, work involved in identifying the students’ tendencies,
datasets design models, and results that were already conduct and the system in which they studied. But, This
exist. There wasn’t any new invention made on the paper considered only academic parameters of the
existing works. students’ not the personal characteristics such as
Anwar & Ahmed [7] presented a new data conduct, past histories, family background in order to
mining application using four machine learning predict if a student is prone to violence.
techniques and feature selection approaches to analyze Eleonora [10], presented the possible
the educational questions asked by the teacher in the interaction between data mining techniques and course
class room with respect to Bloom’s Taxonomy. A management system. The investigator explored several
dataset of pre-classified questions had been collected data mining techniques to deliver most suitable learning
and processed. object of the learner. They emphasized on visualization
Result Analysis had shown the best technique which examined the pictorial representation
performance, particularly with Information Gain term of vast amount of abstract data for comprehending and
selection approach, for cognitively classifying the interpreting.
questions asked by the teacher’s in class room. The They determined, in a given amount of time
results drawn from the study emphasized that the the number of visited learning objects was higher when
application of data mining techniques is domain- student used CMS during school hours, while duration
dependent and therefore it is hard of generalize the of the visit was higher when student used CMS from
findings found in one domain of application to others. their venue. They also classified the CMS materials
Alexandros & Georgios [8], recommended a used at school and home. This endorsed the
framework to record, monitor and examine learner administrator to change the priority of learning objects.
activity behavior while watching or interacting with This investigation endured only with limited attributes
online educational videos. More specifically they that was monitored. Furthermore, they didn’t design
focused on capturing learner performance data, any automation system for the adaptation process.
 IEEE Sponsored 2nd International Conference on Innovations in Information Embedded and Communication
Systems ICIIECS’15

Nutthanon [11], investigated how social media and, it is better than GA and ARM through the
web application helped the learner’s to assist learning theoretic analysis and the experimental results. They
and teaching in classrooms to enhance the performance. hope that some future mining tools would be easier to
They applied both statistics and data mining techniques be used by educators.
on students Facebook activities for data analysis to B. Discussion on Reviewed Papers
determine the effects of Facebook usages for education The above discussion describes that
on students' learning skills. educational data mining has applied extensively in
The results drew that the students who spent numerous applications and offers realistic output in all
more time on education-related posting and educational aspects. The contribution summary of the
commenting earned better grades than the ones who did review is depicted in table 1.
the opposite. They highlighted the learners and teachers In addition to TABLE 1, a diagram is presented from
to feel the advantages and disadvantages of social the ten reviewed studies to which the data mining
media in education. The limitation is, they could have methods are applied in educational setting ranging from
expanded their data collection from various schools to
the year 2009 until 2014. The diagram depicts the
generalize the results.
Dai and Shang [12], presented an approach for number of deployment of various mining technique in
classifying students to predict their final grade the reviewed studies.
TABLE 1: EDM CONTRIBUTIONS SUMMARY
S.N Reference Contribution Technique Algorithm
o
1 [3] Student’s Performance Prediction Classification J 48 , Random Tree

2 [4] Classification Via JRip, J48, EM

Predicting Performance through On-line Discussion Clustering
Forums
3 [5] Decreasing Students Dropout Rate Classification Naïve Bayes

4 [6] Improve Student Retention Rate Classification They used all classification
algorithms
5 [7] Classifying Teacher’s Class Room Questions Classification, Feature Feature Selection:
Selection Term Frequency, Mutual
Information, Information Gain,
and Chi Square.
Classification:
k-Nearest Neighbour, Na'ive
Bayes, Support Vector Machine,
and Rochio Algorithm

6 [8] Analysis behavior with on-line educational videos Visualization Grapviz

7 [9] Student Profiling and grouping Association Rules & A priori Algorithm
Clustering K-Means
8 [10] E-learning System Visualization Weka (Scatter Plot Matrix)
9 [11] Learning through social network Association Rule A priori Algorithm

10 [12] Final Performance prediction in Distance Learning Association Rule Genetic Algorithm

based on features extracted from logged data in an Amongst them, classification was most commonly used
educational web-based system. They combined both (4), next comes the association rule mining (3) followed
association rules mining with genetic algorithm for by visualization (2), clustering (2) and machine learning
determining the most suitable association rules. (1).
They compared the results of the Association
Rule Mining using Genetic Algorithm with the results
 IEEE Sponsored 2nd International C
Conference on Innovations in Information Embedded and Communication
Systems ICIIECS’15

Fig 11. Number of techniques used in the reviewed studies

Besides this investigation there are m many applications Machine learning techniq ques search for an
had been resolved in learning environm ment including, appropriate model which matches with the test data.
• Feedback Evaluation Searching space in machine learn ning is a cognitive
• Recommender System space of n attributes instead of a vector space of n
• Predicting Performance dimensions.
• Student Modeling Database-Oriented Approaches (DO OA)
• Detecting Activity Behavior This approach doesn’t searrch for a best model.
• Classifying or clustering Studeents As an alternative, data model or o database specific
• Social Network participation A Analysis heuristics are used to exploit the features
f of the data.
Association Rule is the classic exaample for extracting
• Schedule Planning
useful data from the database metho ods. In addition, this
• Learning Management System m
approach introduces iterative databaase scanning method
• Dropout & Retention Managem ment to search for frequent item sets in a transactional
• Online Course & E-learning M Management database.
III.COMMON EDM PRACT TICES Neural Network (NN)
Like Data Mining, EDM approaches also NN is a set of interlin nked nodes, called
classified as descriptive or predicttive. Descriptive neurons. It is simple computing dev vice that computes a
approach used to find patterns hidden in large data set function of its inputs which can be outputs of other
that helps in decision making. On the other hand, neurons or attribute values of an n object. In neural
predictive approach constructs modells to predict the network the functional parameters of the neurons are
new class of data. Among the varrious techniques trained by adjusting the connecction to model the
Association Rule Mining and Clusterinng are considered relationship between a set of input attributes
a and output
to be descriptive whereas, rule set, deccision tree, neural attribute.
networks, support vector are connsidered to be Rough Set (RS)
predictive. This approach is a kind of fuzzy membership
A. EDM Techniques set and called as a formal ap pproximation of a
Though EDM has adequate m mining techniques conventional set in terms of a pairr of sets which give
to solve EDM problems; this paper annalyzes the most the lower and the upper approximaation of the original
relevant techniques suitable for educational systems set. Here, a set of objects are arran
nged to form a group
Statistical Analysis (SA) of rough sets which may be used d by DM. Example:
In statistical analysis an opptimal statistical Classification and Clustering.
model is built from fixed training dataa by considering IV. EXPERIMENTA ATION
hypothesis space, rules, patterns, and reegularities. EDM As an experiment, three raandom datasets were
uses many statistical tools, including Bayesian network, chosen from the standard data miningm tool WEKA.
regression, and correlation and cluster aanalyses. These datasets were evaluated d against several
Machine Learning (ML)
 IEEE Sponsored 2nd International Conference on Innovations in Information Embedded and Communication
Systems ICIIECS’15

frequently used classification and clustering algorithms generates the number of clusters as its own, as it is
of Data Mining. The evaluation was got done using shown in TABLE 4. Percent Split is the distribution of
trained dataset. The results were then compared to find instances within the clusters in percentage. The instance
out the most consistent algorithms that could be which does not fall into any cluster is called an outlier
suggested for future researches. The accuracy of the or an abnormal data disparity.
algorithms were assessed and discussed in the next IV. CONCLUSION
section. The names and the description of the datasets At present, educational data mining is paid
are as follows: great attention by most of the researchers, since it
TABLE 2: DATASETS DESCRIPTION evolves in improving the current educational systems.
S. Dataset Instances Attributes This paper concludes that, EDM strongly contributes in
No the betterment of imparting quality in higher education
1 Contact-lenses 24 4
by screening several applications in new dimensions.
2 Credit 1000 20 Though, the achieved results of the reviewed
3 Vote 435 17 works were found to be effective, the real time
implementation of the recommendation systems is still
V. RESULTS AND DISCUSSION diminutive. This investigation establishes, that the
The results were demonstrated in two tables, existing contributions made through EDM focused only
because of the different nature of classification and on higher education. But, the primary education of the
clustering algorithms. The performance of the children is also more important and considered to be the
classification algorithms could be assessed in terms of origin to get succeed in their academic performance. In
prediction accuracy. TABLE 3 describes the prediction future, EDM works may extend in analyzing the
accuracy of decision tree, J48, JRip and Naïve Bayes knowledge process of primary class students

TABLE 3: PREDICTION ACCURACY OF CLASSIFICATION ALGORITHMS

Accuracy(in Percentage)
S.No Classification Algorithm Dataset 1(Contact – Dataset 2(Credit-g) Dataset3(Vote)
Lenses)
1 Decision Tree 100% 100% 99%
2 J48 93% 86% 97%
3 JRip 88% 74% 97%
4 Naïve Bayes 96% 77% 90%

TABLE 4: PERFORMANCE EVALUATION OF CLUSTERING ALGORITHMS

Cluster Methods Dataset 1(Contact –Lenses) Dataset 2(Credit-g) Dataset3(Vote)
S.No NC PS UI NC PS UI NC PS UI
1 EM 2 63%, 37% 0 11 2% , 7%, 5% 0 5 18%, 31% 0
3%, 46%, 14% 29%, 12%
10%, 4%, 1%, 9%
3%, 5%
2 Farthest First 2 46%, 54% 0 2 82%, 18% 0 2 58% , 42% 0
3 Simple K -Means 2 50%, 50% 0 2 64%, 36% 0 2 49%, 51% 0
4 Filtered Clusterer 2 50%, 50% 0 2 64%, 36% 0 2 49%, 51% 0
algorithms with respect to individual datasets. As it is to understand their learning problems at the earliest.
seen, Decision Tree algorithm produced consistent and This would in turn help them in enhancing their quality
good results of 100% in two datasets and 99% in one by giving them the right way of education. This paper is
data set than that of the others. The other three indeed intended to facilitate, both the researchers and
algorithms are inconsistent and their performance varies educators with some useful insights on EDM.
ups and downs from dataset to dataset REFERENCES
On the other hand, the evaluation of clustering 1. Cristóbal Romero, “Educational Data Mining: A
algorithms is quite different and is based on three Review of the State-of-the-Art”, IEEE Transactions On
factors, the number of clusters (NC), Percentage of Systems, Man, And Cybernetics VOL. 10, 2009
Splits (PS) and Unclustered Instances (UI). For any 2. Alejandro Peña-Ayala,” Educational data mining: A
clustering algorithms, the default number of clusters is survey and a data mining-based analysis of recent
2 and can be changed according to the user. However, work”, Expert Systems with Applications 41 (2014)
the algorithm Expectation Maximization (EM) 1432–1462, 2014
 IEEE Sponsored 2nd International Conference on Innovations in Information Embedded and Communication
Systems ICIIECS’15

3. Tripti Mishra, Dr. Dharminder Kumar, Dr. Sangeeta 8. Alexandros Kleftodimos, Georgios Evangelidis, “A
Gupta, ”Mining Students’ Data for Performance framework for recording, monitoring and analyzing
Prediction”, IEEE, Fourth International Conference on learner behavior whilewatching and interacting with
Advanced Computing & Communication Technologies, online educational videos”, IEEE 13th International
2014 Conference on Advanced Learning Technologies, 2013
4. Cristóbal Romero, Manuel-Ignacio López, Jose- 9. Suhem Parack, Zain Zahid, Fatima Merchant, ”
María Luna, Sebastián Ventura,” Predicting students’ Application of Data Mining in Educational Databases
final performance from participation in on-line for Predicting Academic Trends and Patterns”, IEEE
discussion forums”, ELSEVIER, Computers & 13th International Conference on Advanced Learning
Education 68 (2013) 458–472, 2013 Technologies, 2013
5. Dr. Saurabh Pal, “Mining Educational Data Using 10. Eleonora Brtka, Vladimir Brtka, Visnja Ognjenovic
Classification to Decrease Dropout Rate of Students”, and Ivana Berkovic, “The data visualization technique
International Journal Of Multidisciplinary Sciences in e-learning system”, IEEE 10th Jubilee International
And Engineering, VOL. 3, NO. 5, MAY 2012 Symposium on Intelligent Systems and Informatics,
6. Leila Dadkhahan, Mohamad Anas Al Azmeh, 2012
“Critical Appraisal of Data Mining as an Approach to 11. Nutthanon Leelathakul, Kornchawal Chaipah,”
Improve Student Retention Rate”, International Journal Quantitative Effects of using Facebook as Learning
of Engineering and Innovative Technology (IJEIT) Tool on Students’ Performance”, 10th International
Volume 2, Issue 11, May 2013 Joint Conference on Computer Science and Software
7. Anwar Ali Yahya, Addin Osman, Ahmed Abdu Engineering (JCSSE), 2013
Alattab,” Educational Data Mining: A Case Study of 12. Dai Shangping, Zhang Ping, “A Data Mining
Teacher's Classroom Questions”, IEEE, 978-1-4799- Algorithm in Distance Learning”, IEEE, 978 -1-4244-
3516-1113, 2013 1651-6/08, 2008