Educational Data Mining Rule based Recommender Systems
Ghadeer Mobasher
1
, Ahmed Shawish
1,2
and Osman Ibrahim
1
1
Faculty of Informatics and Computer Science, The British University in Egypt, Cairo, Egypt
2
Ain Shams University, Cairo , Egypt
Keywords:
Educational Data Mining, Decision Trees, Reduced Error Pruning and Rule based Recommender System.
Abstract:
Educational Data Mining (EDM) is an emerging multidisciplinary research area, in which data mining tech-
niques are deployed to extract knowledge from educational information systems to help decision makers to
improve the learning process and enhance the academic performance of the students. The available studies
mainly focused on predicting the academic performance based on demographic and study related attributes.
Most of the previous work adopted the decision trees as one of the most famous data mining techniques to
predict rather than extracting real knowledge that reveals the reasons behind student’s dropout. On the other
hand, there were other studies in the psychological track to measure the mental health score based on the
educational environment. This paper proposes a complete EDM framework in a form of a rule based recom-
mender system that is not developed to analyze and predict the student’s performance only, but also to exhibit
the reasons behind it. The proposed framework analyzes the students’ demographic data, study related and
psychological characteristics to extract all possible knowledge from students, teachers and parents.Seeking the
highest possible accuracy in academic performance prediction using a set of powerful data mining techniques.
The framework succeeds to highlight the student’s weak points and provide appropriate recommendations.
The realistic case study that has been conducted on 200 students proves the outstanding performance of the
proposed framework in comparison with the existing ones.
1 INTRODUCTION
Educational Data Mining (EDM)helps decision mak-
ers to improve the learning process and enhance the
academic performance of students in different edu-
cational programs by applying powerful data mining
techniques . The greatest challenge in EDM is to iden-
tify the main factors affecting the student’s academic
performance and hence being able to better predict
the student’s achievement level in a pro-active man-
ner (Prabha and Shanavas, 2014).
Most of the available EDM researches focused
on two categories of data sets as inputs for analysis.
Firstly, the data set related to the student’s educational
related attributes, such as the study and homework
hours , Reading and Writing skills,etc.Second the de-
mographic data set like gender, family related infor-
mation, life style, etc. (Kovacic, 2010).
Despite the enormous researches done based on
these two types of data sets, they only focused on ex-
ploring as much as possible attributes for the purpose
of enhancing the accuracy of their prediction. Their
results were mainly limited to predict the student’s
performance rather than reveling the real reasons be-
hind it (Nasiri and Minaei, 2012).
It is also worth to note that most of the data sets
of these researches were mainly acquired from one
source only, which is the student’s registration forms
without any involvement of other parties like teachers
and parents that participate in the student’s learning
process.
More and above, none of the previous studies in-
cluded the student’s mental health as part of the anal-
ysis. According to the Canadian Mental Health Asso-
ciation, approximately one in five children and youth
has a mental health challenge that directly affects the
student’s learning capabilities. As a result, early iden-
tification is so critical and can lead to improvements
in school and better health outcomes in life (Ramirez,
2014). Regarding the previous educational psychol-
ogy researches that focused on assessing the student’s
mental health score based on the educational settings
using statistical analysis, authors have analyzed the
mental health based on the student’s educational set-
tings, to come out with results that reveal interesting
directly proportional relationships between the stu-
dent’s educational settings of their study program and
their mental health(Li et al., 2008).
292
Mobasher, G., Shawish, A. and Ibrahim, O.
Educational Data Mining Rule based Recommender Systems.
DOI: 10.5220/0006290902920299
In Proceedings of the 9th International Conference on Computer Supported Education (CSEDU 2017) - Volume 1, pages 292-299
ISBN: 978-989-758-239-4
Copyright © 2017 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
We can easily notice the isolation between the two
fields despite their correlation. It’s crucial to add
value to the identification of factors that have great
impact on the student’s academic performance with
the usage of the powerful techniques of data mining
to early alert weak students. Moreover, to provide a
proactive decision making based on the extracted pat-
terns to enhance the student’s learning outcomes.
In this paper we propose a complete framework in
the form of a rule based recommender system, which
is developed not only to analyze the student’s per-
formance, but also to reveal the reasons behind it.In
the proposed framework, we diversify the input data
by combining the student’s demographic data, edu-
cational activities and psychological characteristics.
Moreover, the data set is not only extracted from the
student’s registration forms, but also from the teach-
ers and parents to acquire their evaluation based on
student’s behavior at class and home.
The proposed recommender system will provide a
proactive approach to early alert the issues that might
affect the students academic performance in his/ her
educational life. Moreover, it will provide the student,
parents and the school with the best recommendation,
that points out the weak points of the student that need
to be considered with the appropriate best treatments
based on the corresponding cases. This paper high-
lights three main contributions as follows:
Concluding the most relevant attributes from the
previous researches that have the highest influ-
ence on the student’s academic performance.
Deploy a set of very powerful data mining
techniques to accurately predict the student’s
academic performance and extract all possible
knowledge from the input data.
Develop a recommender system to provide the
student, parents and teachers with the appropri-
ate recommendation based on the corresponding
problems.
An extensive simulation studies have been con-
ducted on a realistic real cases including 200 Elemen-
tary Students, revealing the outstanding performance
of the proposed framework in comparison with the
previous ones.
This paper is organized as follows: Section 2 illus-
trates the background. Section 3 introduces the pro-
posed techniques and methodologies to develop the
EDM Rule Based Recommender System. Section 4
describes the experiment and the results. Finally, Sec-
tion 5 discusses the conclusion and the future work.
2 RELATED WORK
In this section we provide a comprehensive study of
all the previous EDM Researches and we also review
the available educational recommender systems.
2.1 Previous Predictive Models
The previous predictive models only focused on us-
ing the student’s demographic data like gender, age,
family status, family income and qualifications. In
addition to the study related attributes including the
homework and study hours as well as the previous
achievements and grades. Bhardwaj and pal predicted
the student’s performance at the end of the semester
through student’s data such as attendance, assign-
ments marks and class test marks (Baradwaj and Pal,
2012). From diverse literature, the observed predicted
poor performance has been mostly traced to poor pre-
vious scores, demographic data as well as the level of
intelligence(Ahmed and Elaraby, 2014). In addition
to other background factors such as socioeconomic
status, family education and occupation, religion and
even ethnicity have been identified as factors that have
highest implications on the student’s academic per-
formance(Berger and Archer, 2016). These previous
work were only limited to provide the prediction of
the academic success or failure, without illustrating
the reasons of this prediction. Most of the previous
researches have focused to gather more than 40 at-
tributes in their data set to predict the student’s aca-
demic performance. These attributes were from the
same type of data category whether demographic ,
study related attributes or both , that lead to lack of
diversity of predicting rules. As a result, these gener-
ated rules did not fully extract the knowledge for the
reasons behind the student’s dropout.
Apart from the previously mentioned work, there
were previous statistical analysis models from the per-
spective of educational psychology that conducted a
couple of studies to examine the correlation between
the mental health and the academic performance. Pre-
vious researches studied the academic performance
of students with respect to the correlation with their
mental health behavior. Their results demonstrate
that feelings of anxiety, depression, and time pres-
sure negatively affected the performance of these stu-
dents. However, participating in extracurricular ac-
tivities alongside having a good support system pos-
itively affected the academic performance(McLeod
et al., 2012). Moreover, The Center for Addiction and
Mental Health found through surveys that final year
students were least likely to report these symptoms as
compared to students in other years (Weare and Nind,
Educational Data Mining Rule based Recommender Systems
293
2011).
After surveying the previous related work, we
conclude that the previous attempt’s main goal was
the identification of the highest impact factors on aca-
demic performance. Some of them were from the ed-
ucational data mining perspective , while others from
the educational psychology methodology . Both fields
have the same goal, but each of them have different
methodology .
2.2 Previous Educational Data Mining
Recommender Systems
Concerning the previous recommender systems, most
of them are student oriented and they predict the per-
formance of a specific subject or provide a single
learning resource rather than predicting student’s per-
formance in each course and providing overall pre-
diction to the student’s final score. For example,
Piedade (Piedade and Santos, 2008) that proposed a
SRM model provide support in the form of an ef-
fective student institution relationship through mon-
itoring of the students and their academic activities.
Also,Kanokwan (KuyoroShade et al., 2013) built on
the previously mentioned SRM model to propose
an intelligent recommendation system framework for
student relationship management that can assess the
performance of the students and provide the appropri-
ate recommendation for the choice of single course.
The existing systems mostly focused on predict-
ing the student’s performance in a particular course
with respect to the student’s data from the admission .
Student’s oriented systems have the purpose of aiding
students to the best decision for specific course regis-
tration enrollment. The type of the recommendations
was too brief, they missed illustrating the methodolo-
gies to apply them.
We believe that the real recommender system
should provide the suitable advises to enhance the
student’s overall academic performance based on the
type of the challenges that the student is suffering
from during the academic year.
3 PROPOSED FRAMEWORK
This section describes the proposed framework
through four sub sections. First, the new combination
of input data set is explained. As well as, detailed il-
lustration of data pre-processing is presented. Second
subsection describes the data analysis and processing.
Finally, the implementation details of the proposed
framework is discussed.
3.1 Input Data Model
The proposed framework firstly focuses on merg-
ing the demographic and study related attributes with
the educational psychology fields, by adding the stu-
dent’s psychological characteristics to the previously
used data set (i.e., the students’ demographic data
and study related ones). After Surveying the previ-
ously used factors for predicting the student’s aca-
demic performance, we picked the most relevant
attributes based on their rationale and correlation
with the academic performance (Osadan and Bur-
rage, 2013) (Cimmiyotti, 2013). Concerning the stu-
dent’s psychological characteristics, they were ex-
tracted through group of symptoms that asses the stu-
dent’s mental health based on children fact sheet from
National Alliance on Mental Health (NAMI) (Deb
et al., 2015). According to NAMI, the top 5 men-
tal illness that the school student’s might suffer from
are Anxiety, Hyperactivity, Bipolar, Conduct, and Re-
active Attachment Disorders. The previous studies
proved that low mental health score has negative aca-
demic performance implications
To remove any unnecessary attribute, we conduct
a feature selection analysis using Information Gain as
an attribute evaluator and Ranker as a search method.
The feature selection analysis dropped 8 attributes
from the data set, that includes 45 attributes as they
report zero gain information gain.
Three questionnaires have been designed to col-
lect the above listed data for the student, parents and
teachers. The first one was questioned to the stu-
dent, mainly focusing on the set of pre-mentioned de-
mographic data. The other two questionnaires, were
targeting parents and teacher’s evaluation for the stu-
dent’s psychological characteristics and study related
skills.
The input model is divided into three main cate-
gories, the first is the demographic data which is il-
lustrated in Table 1. The second category is the study
related attributes illustrated in Table 2. Finally, the
third category is psychological characteristics shown
in Table 3.
3.2 Data Analysis
This paper proposes a hybrid system which is based
on data mining techniques using classification algo-
rithms, specially focusing on four rule induction al-
gorithms One R, Zero R, JRIP and PART. And four
decision algorithms J48, Random tree, REP tree and
Decision stump. The first four classifiers belong to
Rules and the second ones belong to Decision Trees.
These classification algorithms belong to the ”white
CSEDU 2017 - 9th International Conference on Computer Supported Education
294
Table 1: Demographic Arributes.
Attribute Name Expected Values
Stage 1,2,3,4,5,6 Primary
Gender Male, female
Home Type Flat, villa
Father qualification
Mother qualification
School level,
BSc.
MSc
PHD
Family income
Poor,
Middle,
High
Parental status
Married,
Divorced,
Separated.
No. of siblings Numeric
Table 2: Study Related Attributes.
Variable Name Variable format
Number of courses
NumericHomework hours
Study hours
Counting skills
Poor,
Average,
Excellent
Arithmetic knowledge
Reading skills
Writing skills
Hand-writing skills
box” classification model and can be used directly for
decision making. To validate that our proposed work
outperform the previous related work, after merging
between the demographic, study related and psycho-
logical characteristics and after applying 8 different
powerful data mining techniques to predict the aca-
demic performance.
In order to choose the best data mining technique,
there are three main concerns for the model evalua-
tion : the tree size, tree complexity and prediction ac-
curacy of the decision tree.
The proposal aims to analyze student’s demo-
graphic data, study related details and psychological
characteristics in terms of final state to figure whether
the student is on the right track or struggling or even
failing. In addition to extensive comparison of our
proposed model with the other previous related mod-
els.
3.3 Framework Implementation
The proposed recommender system is implemented
with java language using Net Beans version 7.3 as a
java environment. WEKA Java API is be used to im-
plement the data mining and machine learning algo-
rithm. The intended user is the Data Analyst. The
recommender system enables the data analyst to up-
load the data set and then view the predicted set of
Table 3: Psychological Attributes.
Variable Name Variable
format
Talkative
Yes/No
Lack of motivation
lack of organization
History of frequent suspension
Mood swinging
Frequent Abscence
Failure to finish work
Bossy
Easily distracted
Always,
Never,
Often
disability to listen
Blurt Out
Trouble playing quietly
Makes a lot of mistakes
Lying
Physical fights
Stealing
Unusual speech patterns
Toileting issues
Refusal to answer simple questions
Memory abilities Poor,
Average,
Excellent.
Attention abilities
Eye-contact abilities
rules and the generated tree. Moreover, the data ana-
lyst can add student’s information to predict the over-
all academic performance, the predicted value will
be displayed. The recommender system isn’t only
limited to the prediction of state, but also provides
the student, parents and teachers with the correspond-
ing recommendations and strategies to follow to im-
prove the learning outcomes. These recommenda-
tions are based on experimented studies for enhancing
the student’s academic performance. In addition to
the mentioned above functionalities, the System will
also alert all parties with the possible upcoming men-
tal illnesses that the student might suffer from. Due
to the gathering of the psychological characteristics
that formulate set of symptoms to couple of mental
illnesses that might negatively impact the academic
performance, as mentioned in section 2.
4 CASE STUDY
This section presents the realistic case study and dis-
cusses its results. The Student’s academic perfor-
mance prediction and recommendations will be illus-
trated with a comparative analysis between our pro-
posed framework and the previous related work.
Educational Data Mining Rule based Recommender Systems
295
4.1 Experiment Settings
Once the questionnaires have been acquired from the
pre-mentioned sources and after the data set passed
through the pre-processing phase. The experiment is
conducted on 200 elementary students , after elimi-
nating the inconsistencies between Teachers and Par-
ent’s Response on the study related data and psycho-
logical characteristics.
The data is then analyzed for knowledge extrac-
tion; the data is first classified on the hybrid 8 clas-
sifier’s algorithms using 10- fold cross validation
method. Data were randomly divided into 10 parts.
Each held holds the next and learning scheme trained
nine-tenth of the rest of the data set, then the error rate
is calculated in the holdout set. The learning proce-
dure is performed 10 times on different random train-
ing set. Finally, an average of 10 error estimates to
produce estimates of the overall error. Detailed result
analysis will be discussed in the coming subsection.
4.2 Predictive Model Results and
Validation
In terms of prediction accuracy, REP Tree outperform
with a prediction accuracy of 73.6 %. While the least
prediction was for Zero R with 45.5 % as prediction
accuracy. Detailed bar chart graph is illustrated in fig-
ure 1.
Figure 1: Classifier’s Prediction Accuracy.
Regarding the classifier’s tree size, REP tree has
the smallest tree size. While, Random Tree has the
largest tree size. The complexity of the tree is mea-
sured by the size of the tree, the total number of leaves
and the tree’s depth. The less the tree size, total leaves
and tree depth the less complicated the tree. In figure
2, detailed comparison between classifier’s tree size
and complexity.
Although Zero R has the least tree size and com-
plexity, however it is the least reliable classifier to ap-
ply, because the extracted rules lost structural com-
plexity. Only one rule is generated to predict the aca-
demic performance.
Figure 2: Classifier’s tree size and complexity.
It is very clear to conclude that REP tree is the
best classifier to apply.As the size of the REP tree
is the smallest with the lowest complexity and high-
est prediction accuracy. REP tree is a post-pruning
method; this decision tree finds the smallest version
of the most accurate sub-tree with respect to the prun-
ing set. During the pruning the size of the tree will
get reduced and reduces the complexity and thus im-
proves the prediction accuracy. Regarding the Rep’s
prediction accuracy results analysis 73.5 % were pre-
dicted true when they were actually true, while 12.5 %
were predicted true however, they were actually false.
In Table 4 , the generated REP Decision Tree.
Table 4: Generated Rep Rules.
Generated Rules Predicted State
If the Arithmetic Knowl-
edge == Excellent
Then the state == On the
Right Track
If the Arithmetic Knowl-
edge == Average and Read-
ing Skills==Excellent ==
Excellent
Then the state == Strug-
gling
If the Arithmetic Knowl-
edge == Average and Read-
ing Skills==Poor
Then the state == Fail-
ing
If the Arithmetic Knowl-
edge == Average and Read-
ing Skills== Average and
History of Frequent Suspen-
sion==Yes
Then the state == Fail-
ing
If the Arithmetic Knowl-
edge == Average and Read-
ing Skills== Average and
History of Frequent Suspen-
sion==No
Then the state == Strug-
gling
If the Arithmetic Knowl-
edge == Poor and Makes
lots of Mistakes ==Always
Then the state == Fail-
ing
If the Arithmetic Knowl-
edge == Poor and Makes
lots of Mistakes ==Never
Then the state == Strug-
gling
If the Arithmetic Knowl-
edge == Poor and Makes
lots of Mistakes ==Often
Then the state == Strug-
gling
CSEDU 2017 - 9th International Conference on Computer Supported Education
296
If we applied the previous related research strat-
egy that used the demographic data to predict the aca-
demic performance on our data set. The best clas-
sifier is J48 with a correctly classified instances and
the predicted rules mainly depend on the fathers and
mother’s qualifications, home type, family income
and frequent absence. While if we used the demo-
graphic data and the study related attributes to predict
the academic performance, the best classifier is the
REP tree. Arithmetic knowledge, Number of courses,
Writing Skills, Homework Hours and stage have the
highest impact on the prediction of performance.
Finally, if we applied the third strategy of using
the psychological characteristics to predict the perfor-
mance, the best classifier was J48 correctly classified
instances. History of frequent suspension, Failure to
finish work, blurt out, Memory Abilities, Lying and
disability to listen were mainly considered in the pre-
diction of performance.
In figure 3, detailed comparative prediction accu-
racy between old strategies vs our work.
Figure 3: Comparison between previous work and our
model’s prediction accuracy.
Beyond the out performance of our proposed strat-
egy in terms of accuracy percentage and our extracted
predictive rules , which are consistent with the theo-
retical aspects of the educational principles. We can
assert that our work added great values to the predic-
tion of the academic performance. After applying our
proposed methodology to merge between educational
data mining and educational psychology perspectives,
new results have been generated that better predicted
the student’s academic performance.
4.3 Recommeder System Discussion
The data analyst first logs in and browses for the file to
upload it. Then the system automatically applies the
8 classifiers and uses the highest best classifier’s pre-
diction accuracy, which is REP Tree in our case. Then
the data analyst can have the option to see the clas-
sifier’s prediction rules and generated decision tree
as well. Also a detailed message is displayed with
detailed classification’s accuracy. In addition to the
functionalities mentioned above, the data analyst can
enter the data of the student who wants to predict
his/her overall academic performance. After the an-
alyst fills in the student’s information, the system dis-
plays the predicted academic state of the student and
presents the appropriate recommendations for the stu-
dent, the parents and the teacher.
Moreover, it alerts all parties with the mental dis-
order that the student might face in the future. It is
possible that the student might not face any academic
problems currently, but in the future, the student will
be facing academic performance challenges due the
mental illness disorder. So the system will not only
provide the analyst with the predicted academic state
and the appropriate recommendations, but highlights
any upcoming mental illness that the student might
shown in the future. Snippets of the system’s inter-
face are illustrated in figure 4, which presents an ex-
ample of the parent’s , teacher’s and student’s recom-
mendation after entering student’s data and predict-
ing performance. The proposed system avoided the
Figure 4: Screenshot of framework interface .
drawbacks of other system. It provides recommenda-
tion for all parties that participate in the student’s ed-
ucational life, including the student, the teachers and
parents. Samples of the recommendations addressed
to the student are illustrated in table 5. Table 6 il-
lustrates sample of recommendations for the teachers
and parents.
The extracted recommendations are based on
proven classroom fact sheets (for Disease Control
et al., 2013) (Hite, 2009) (De Corte et al., ).
Educational Data Mining Rule based Recommender Systems
297
Table 5: Samples of the suggested recommendations for
Students.
Problem Suggested Recommendation
The student’s arith-
metic knowledge
and reading skills
are average or poor
Play arithmetic games so
that you can have fun and
at the same time learn and
practice.
Write the small steps as a
draft while solving rather
than memorizing the num-
bers to avoid mistakes.
Scan before you read.
Practice, the more you read
the better reader you will
become and smarter too, so
feed your mind.
The student is
suffering from
AD/HD ,(s)he is
easily distracted
,blurt out ,poor at-
tention and memory
abilities,talkative
,lacks organization
, having trouble
to play quietly
and makes lots of
mistakes
Enhance your social skills
and try to get involved with
other peers.
When you feel low or sad,
try to divide your stud-
ies into small manageable
parts.
Moreover, most of the previous attempts were fo-
cusing on predicting the student’s performance in a
specific course with respect to the student’s informa-
tion gathered from the admission. It only focuses on
identifying the factors that have high impact on the
student’s success or failure in certain subject. Based
on the no-free theorem, the system that was able to
predict students performance in a particular course
may not do so in the overall performance of the stu-
dents at the end of each academic year.
One of the previous recommender system at-
tempts, let the user add the IF-ELSE rules manually
and this of course increases the human error. while
the implemented system let the data analyst loads the
data set, enabling dynamic prediction rules extraction,
even if the data set changes.
5 CONCLUSION AND FUTURE
WORK
This paper proposes a complete EDM framework in
the form of a rule based recommender system that
analyze the student’s academic performance to point
out the student’s weak points and provide appropriate
Table 6: Samples of the suggested recommendations for
Parents and Teacher.
Problem Suggested Recommendation
The student’s arith-
metic knowledge
and reading skills
are average or poor
Implement a coherent read-
ing program level.
Focus on fluency and com-
prehension.
Create a culture that encour-
ages learning, thinking, re-
flection and self-analysis
The student is
suffering from
AD/HD ,(s)he is
easily distracted
,blurt out ,poor at-
tention and memory
abilities,talkative
,lacks organization
, having trouble
to play quietly
and makes lots of
mistakes
Provide the student with
recorded books as an alter-
native to self-reading when
the student’s concentration
is low.
Break assigned reading
into manageable segments
and monitor the student’s
progress, checking compre-
hension periodically.
Devise a flexible curricu-
lum that accommodates the
sometimes rapid changes in
the student’s ability to per-
form consistently in school.
When energy is low, reduce
academic demands; when
energy is high, increase op-
portunities for achievement.
Identify a place where the
student can go for privacy
until he/she regains self-
control.
recommendations for treatment.The paper combined
three major factors that affect the student’s academic
performance: demographic data, educational related
attributes and psychological characteristics. The real-
istic case-study conducted on 200 students assured the
outstanding capabilities of the academic performance
prediction, depth of knowledge extraction, and great
benefit of recommendations provided by the proposed
framework in comparison with the available work.
The generated rules also showed a perfect matching
with the scientific proved facts. The diversification of
the data sources and the involvement of teachers and
parents present a notable improvement in this work.
Regarding the future work, a larger data set
needed to be used in different academic stages. More-
over, new psychological characteristics need to be
added with the supervision of professional psychol-
ogist to better discover new patterns and enhance the
prediction results.
CSEDU 2017 - 9th International Conference on Computer Supported Education
298
REFERENCES
Ahmed, A. B. E. D. and Elaraby, I. S. (2014). Data mining:
A prediction for student’s performance using classifi-
cation method. World Journal of Computer Applica-
tion and Technology, 2(2):43–47.
Baradwaj, B. K. and Pal, S. (2012). Mining educational
data to analyze students’ performance. arXiv preprint
arXiv:1201.3417.
Berger, N. and Archer, J. (2016). School socio-economic
status and student socio-academic achievement goals
in upper secondary contexts. Social Psychology of Ed-
ucation, 19(1):175–194.
Cimmiyotti, C. B. (2013). Impact of reading ability on aca-
demic performance at the primary level.
De Corte, E., Walberg, H., Fraser, B., Kirst, M., Teichler,
U., and Wang, M. The international academy of edu-
cation.
Deb, S., Strodl, E., and Sun, J. (2015). Academic stress,
parental pressure, anxiety and mental health among
indian high school students. International Journal of
Psychology and Behavioral Sciences, 5(1):26–34.
for Disease Control, C. et al. (2013). Make a difference at
your school.
Hite, S. (2009). Improving problem solving by improving
reading skills.
Kovacic, Z. (2010). Early prediction of student success:
Mining students’ enrolment data.
KuyoroShade, O., Oludele, A., Okolie Samuel, O., and
Nicolae, G. (2013). Framework of recommendation
system for tertiary. Framework, 2(04).
Li, H., Li, W., Liu, Q., Zhao, A., Prevatt, F., and Yang, J.
(2008). Variables predicting the mental health status
of chinese college students. Asian Journal of Psychi-
atry, 1(2):37–41.
McLeod, J. D., Uemura, R., and Rohrman, S. (2012). Ado-
lescent mental health, behavior problems, and aca-
demic achievement. Journal of health and social be-
havior, page 0022146512462888.
Nasiri, M. and Minaei, B. (2012). Predicting gpa and aca-
demic dismissal in lms using educational data mining:
A case mining. In 6th National and 3rd International
conference of e-Learning and e-Teaching, pages 53–
58. IEEE.
Osadan, R. and Burrage, I. A. (2013). The effect of age,
societal status and sexuality on students elementary
schooling.
Piedade, M. B. and Santos, M. Y. (2008). Student relation-
ship management: Concept, practice and technologi-
cal support. In 2008 IEEE International Engineering
Management Conference, pages 1–5. IEEE.
Prabha, S. L. and Shanavas, D. A. M. (2014). Educational
data mining applications. Operations Research and
Applications: An International Journal (ORAJ), 1(1).
Ramirez, J. (2014). The relationship between school-based
mental health services and academic achievement.
Weare, K. and Nind, M. (2011). Mental health promotion
and problem prevention in schools: what does the evi-
dence say? Health promotion international, 26(suppl
1):i29–i69.
Educational Data Mining Rule based Recommender Systems
299