You are on page 1of 7

Imperial Journal of Interdisciplinary Research (IJIR)

Vol-2, Issue-12, 2016


ISSN: 2454-1362, http://www.onlinejournal.in

Recognition of Slow Learners Using


Classification Data Mining Techniques
Mukesh Kumar1, Shankar Shambhu2 & Punam Aggarwal3
1,2
Chitkara University, HP (INDIA)
3
Smt. Aruna Asaf Ali Govt. P. G. College, Kalka (INDIA)

Abstract: Educational Data Mining is used to predict Clementine, Intelligent Miner etc with the proper
the future learning behavior of the student. It is still interface for all techniques of mining.
a research topic for the researcher who wants do
better result from the prediction of the student. The Different research gives their own definition of
results of all these techniques help the teachers, education, but the overall result of each research is
management, and administrator to draft new rules same i.e. How to improve the overall education
and policy for the improvement of the educational system? There are lots of factors which affect the
standards and hence overall results and student education system like the student, management,
retention. Taking this point in mind work has been administrators, infrastructures, teachers, teaching
done to find the slow learner in a High School class methodology, basis facilities for boys and girls (like
and then provide timely help to them for improving separate toilets for M/F), transportation facilities etc.
their overall result. There are lots of techniques of So with the help of educational data mining
data mining are available for use but we are techniques, detailed analyses are performed on these
selecting only those techniques which are mostly factors and find out those factors which affect the
used by different research for their result prediction education of the student and hence students
like J48, REPTree, Naive Bayes, SMO, Multilayer underperformed.
Perceptron. On the collected dataset Multilayer
Perception classification algorithm gives 87.43% 2. Liability of data mining in academics
accuracy when using whole dataset as training setting:
dataset and SMO and J48 gives 69.00% accuracy
when using 10-fold cross validation algorithm. As already written EDM play a significant role for
the overall development of the education? With the
Keywords: Data Mining, Educational DM, help of this following question can be answered
Prediction, Classification, Clustering. which are enlisted below:

1. Introduction I. Who is the weak student in a particular class?


II. Who is likely being the dropout in the
Educational data mining is one of the applications education?
of data mining. Data mining is used to find the III. Which subject students like most in their overall
hidden pattern from a huge dataset and then apply course?
that hidden patterns for the decision making in IV. Which courses most attract the student in
future. Its application is not limited to education but education?
also covers fields like sales, retail, transportation, V. Find out the possible attribute which effect the
sports, marketing etc. In education, these data mining student education and hence performance?
techniques are used to predict slow learners, dropout, VI. How we can help those student who are slow
under-performer etc and hence provide timely help to learner, under-performer and problem of
those students who are the problem in education. education dropout.
EDM is also broadly applied to E-learning system, VII. Most importantly predicting result of the student
classroom teaching, MOOC Course learning, in final examination.
curriculum redesign analysis, distance education etc.
It is a process to find some knowledge of the 3. Proposed work for this research
database and then apply that knowledge for future
improvement. It is also known as KDD (Knowledge Education plays a crucial role in the development of
discovery in a database). There is so many tools are the society. If education system and technology work
available in the commercial market for the mining together then it makes unbelievable growth for
purpose like RapidMiner, WEKA, DBMiner,

Imperial Journal of Interdisciplinary Research (IJIR) Page 741


Imperial Journal of Interdisciplinary Research (IJIR)
Vol-2, Issue-12, 2016
ISSN: 2454-1362, http://www.onlinejournal.in

Fig 1. Use of Data Mining in Education setting

the society. At present technology are also used in IV. Observe those variables which are extremely
education like E-learning, MOOC Courses, Smart influencing for the prediction of the student
Classes are introduced in school and it really works academics performance.
well for the overall development of the student. V. At the end compare the predictive result of all
Technologies like data mining are also introduced in the techniques and choose the best classification
the education sector for the prediction of the student algorithm result for further improvement.
in their future learning behaviors. So taken these
prospective in mind, work has been done to predict At the end of this paper, all these motives are
the slow learners in a class and hence provide timely fulfilled and a brief conclusion is given. It helps for
help for improving their final result. The major further research in this novel field of data mining.
motives behind this work are:
4. Literature Survey: Background and
I. Find source of data collection for creating prior work in this area
dataset which contains predictive variables

II. Selecting best data mining technologies for the Use of data mining in education is tremendous.
analysis of the student performance. But still, lots of researchers are working on
III. After analysis of the dataset identifies those educational data mining techniques for the
students who are slow learners and need betterment of education. As already told it's a broad
immediate help in their study.
field and not limited to the present discussion like the
prediction of slow learners in a class.

Fig 2: Data Mining Process to be taken under consideration

Han Jiawei and Micheline Kamber, Education Integration, Selection, Transformation of data and
Data Mining is a process of Knowledge Discovery in pattern Evaluation Phases [1].
a huge database which consists of Cleaning,

Imperial Journal of Interdisciplinary Research (IJIR) Page 742


Imperial Journal of Interdisciplinary Research (IJIR)
Vol-2, Issue-12, 2016
ISSN: 2454-1362, http://www.onlinejournal.in

S Weiss et al, They explain data mining as search different variables which are used for the prediction.
techniques, which search for the valuable They used seven different class predictor variables
information for a huge database and hence apply that for their experimentation [10].
information for better decision making [2]. Lars Schmidt-Thieme et al, they applied machine
William J. Frawley et al and Tech. Forecast, Data learning algorithm for prediction, a result of which
Mining is a detailed process of extracting useful was further used to improve the academic
pattern or useful information which is formerly performance of the student. To deal with the problem
unknown to the database user. The known pattern or of the imbalanced data they applied three different
information may include information like association methods and hence found satisfactory results. After
rules between variables, pattern finding between balanced the datasets they further used SVM for a
variables etc. [3] [4]. small dataset and Decision tree algorithm for larger
S. Pal et al, they are using linear regression dataset [11].
technique for their analysis and find that factors like V. Ramesh et al, they applied survey
mother’s education and family income of students methodology to make the final dataset with some
affect their academic performance [5]. significant variable of students and with
M Bala, Dr. DB Ojha, They define EDM as experimental methodology tried to found only those
techniques which are helpful in finding the unknown variables which influencing the final result of the
facts from a larger database, which are impossible to student. They applied SMO, J48, REPTree, Naive
find manually and hence that information, are Bayes and Multilayer perception techniques for their
effectively used in the education setting. It is used to experimentation purpose. After analysis, they found
increase the student retention rate, improve the that factor like parent’s occupation plays a very
educational standard, and help administrator for important role in student performance [12].
setting new rules and regulation for improving Applying EDM techniques for knowledge
educational standard [6]. discovery is important for the teachers, management,
Ying Zhang et al, to extract useful information, and student. They all are using this knowledge for
data mining used to combine the machine learning, the improvement of the education system. Teachers
visualization, and statistical approaches. There are are using this knowledge for improving their
lots of techniques are adopted to collect the data for teaching standard and the student is using to
making dataset for analysis like questionnaires, improving their learning skill. Management of the
feedback form, interview, discussion. After institution is using this knowledge for improving
collecting all data make a dataset according to the infrastructure standard, provide basic facilities to the
selected tool for analysis and then apply some student and decision making.
techniques for analysis like Classification,
Clustering, Linear Regression, Support Vector 5. Data collection and proposed
Machine, Decision Tree, Naive Bayes and K-mean. methodology
Student learning behaviors, Course learning, student
retention rate, course suitability etc are predicted By Han Jiawei, Micheline Kamber, EDM
using data mining techniques [7]. software's should be developed in such way that the
Cortez and Silva, taken twenty-nine attributes for users can analyze the student data with different
prediction of the result in Mathematics and dimensions, enables to categorize and summarize the
Portuguese. They applied data mining algorithm (like desired results [2].
Decision Tree, Neural Network, Support Vector To complete this work a survey was conducted on
Machine and Random forest) on the dataset of 788 student and then for analysis purpose a data mining
students of two schools from the Alentejo region of tool should be used. Here WEKA tool is used for the
Portugal. After analysis, they found that Decision analysis purpose because it is open source software
Tree (DT) and Neural Network (NN) had 93% and and almost all the data mining techniques are
91% accuracy in predicting the result according to implemented on it. After the detailed survey and
two- class (pass/fail) respectively [8]. having the discussion with the experts, some
Galit.et.al, in his case study they analyze the attributes are selected related to the students which
student’s data for predicting their future learning are mostly affecting the academic performance in
behaviour and hence the result. They also predict the high schools. These attributes are also known as
student result and warn them that they are at risk of input variables for the analysis of the dataset.
failure in final examination and provide timely help
to them [9].
M. Ramaswami et al, for the analysis of the The data are collected from two different high
educational outcome of the student’s in higher schools with the help of survey method and after that
secondary education they used CHAID prediction put it in desired file format required for the analysis.
techniques to find the interrelationship between

Imperial Journal of Interdisciplinary Research (IJIR) Page 743


Imperial Journal of Interdisciplinary Research (IJIR)
Vol-2, Issue-12, 2016
ISSN: 2454-1362, http://www.onlinejournal.in

Table-1: Selected Attributes of student taken for analysis purpose


S. NO ATTRIBUTE DESCRIPTION DOMAIN VALUES OF THE
ATTRIBUTES
1 TY_HS Types of High School { Govt, Private, Govt_add}

2 TY_BRD Types of Education board {State Board, CBSE, ICSE}

3 MED_INS Medium of Instruction {Hindi, English, Pahari}

4 TY_SCL Type of School {Boys, Girls, Co-education}

5 GND Gender of student {Girls, Boys}

6 PRI_TUT Private Tuition taken {Yes, No}

7 AR_SCL Location of the school area {Urban, Rural}

8 INT_GRD Internal Grade of student {A, B, C, D, E, F}

9 MOB Mobile Phone {Yes, No}

10 COM Computer at Home {Yes, No}

11 INT_ACC Internet access to student {Yes, No}

12 ATND Attendance in the school { In % age out of attendance taken}

13 CLASS Eligible of Not Eligible {E, NE}

Here for the implementation purpose CSV file School, types of education board, medium of
format are used for WEKA tool. WEKA is a open instruction and type of school.
source software tool kit and support maximum of Find out the High Potential Attribute: After
classification, clustering and association rule pre-processing of dataset find the high potential
algorithms attribute which are critically affects the overall
dataset with different attribute selection method. In
6. Implementation of EDM techniques WEKA the different attribute evaluator algorithm are
on dataset CorrelationAttributeEval, GainRatioAttributeEval,
InfoGainAttributeEval, OneRAttributeEval,
During this phase of work, first of all we pre- PrincipalComponents, ReliefAttributeEval, and
process our dataset with the help of WEKA SymmetricalUncertAttributeEval which are further
Preprocess feature on the tool interface. used different search method like BestFirst,
Apply Filters on the dataset: For pre-processing GreedyStepwise and Ranker. In this work all the
of dataset, implement filters on it to remove those dataset are used as a training dataset and we are not
attributes who are not supposed to be significant for using 10-fold cross validation method because of less
the result prediction. After implementation of filter, data in dataset. After applying all algorithm of
dataset left with only eight different attributes along attribute evaluator INT_GRD, INT_ACC and ATND
with class. The removed attributes are types of High are found most important
for the classification of the dataset with two class dataset. In table below the entire algorithm with their
values. Most of the algorithms are used ranker search search method and first rank attribute are mentioned.
for find the high potential attributes in the given

Imperial Journal of Interdisciplinary Research (IJIR) Page 744


Imperial Journal of Interdisciplinary Research (IJIR)
Vol-2, Issue-12, 2016
ISSN: 2454-1362, http://www.onlinejournal.in

Table-2: High Potential attributes selection form the dataset


ATTRIBUTE EVALUATOR SEARCH METHOD FIRST RANK ATTRIBUTE

cfsSubsetEval GreedyStepwise INT_GRD


CorrelationAttributeEval Ranker INT_ACC
GainRatioAttributeEval Ranker INT_GRD
InfoGainAttributeEval Ranker INT_GRD
OneRAttributeEval Ranker ATND
ReliefFAttributeEval Ranker INT_GRD
SymmetricalUncertAttributeEval Ranker, BestFirst INT_GRD

At the end of this section, it is clear that only classification algorithm like Naive Bayes, SMO, J48,
seven attributes in the dataset are useful for the REPTree and Multilayer Perceptron. These entire
prediction of the class attribute and rest of the algorithms are also tested with 10 fold cross
attributes are not affecting the overall result of the validation check as well as using full training data
analysis. set. The Correctly and Incorrectly classified
Instances after implementing listed algorithm under
7. Results of implementation 10-fold validation are given in table below:

After the completed the pre-processing task,


dataset is tested and analyzed with five well known

Table-3: Correctly & Incorrectly Classified Instances using 10 fold validation check
Data Mining Techniques Correctly Classified Incorrectly Classified
used Instances Instances
Multilayer Perceptron 57.2864 % 42.7136 %
Naive Bayes 67.3367 % 32.6633 %
J48 69.3467 % 30.6533 %
SMO 69.3467 % 30.6533 %
REPTree 67.8392 % 32.1608 %
OneR 67.8392 % 32.1608 %
ZeroR 69.3467 % 30.6533 %

Using 10-fold validation check classification correctly classified instances are 69.3 percent. Which
algorithm like J48, SMO and ZeroR are performing is acceptable as the baseline condition given by
better than other algorithm under consideration. The ZeroR algorithm is also 69.3 percent?

Fig - 3: Comparison of classifiation accuracy with the help of graph

Imperial Journal of Interdisciplinary Research (IJIR) Page 745


Imperial Journal of Interdisciplinary Research (IJIR)
Vol-2, Issue-12, 2016
ISSN: 2454-1362, http://www.onlinejournal.in

The Correctly and Incorrectly classified Instances as training dataset are given in table below:
after implementing listed algorithm using full dataset

Table-4: Correctly & Incorrectly Classified Instances using training data set
Data Mining Techniques Correctly Classified Incorrectly Classified
used Instances Instances
Multilayer Perceptron 87.4372 % 12.5628 %
Naive Bayes 69.3467 % 30.6533 %
J48 69.3467 % 30.6533 %
SMO 69.3467 % 30.6533 %
REPTree 72.3618 % 27.6382 %
OneR 69.3467 % 30.6533 %
ZeroR 69.3467 % 30.6533 %

Using full dataset as training dataset for are performing exceptional well with 87.43 percent
classification algorithm like Multilayer Perceptron correctly classified instances.

Fig - 3: Comparison of classification accuracy with the help of graph


8. Conclusion and future scope medium of instruction. May be these attribute also
affect the performance of the student in education.
There are lot of drawbacks in education system
like midterm evaluation system use. It is really not
understood that why midterm evaluation is taken 9. References
throughout the year. In this particular paper,
[1]Han Jiawei, Micheline Kamber, “Data Mining:
classification algorithms are applied on dataset of
Concepts and Technique”. Morgan Kaufmannv Publishers,
students, to predict slow learners in a class. For that a 2000.
model was created based selected attribute of
student. From so many classifiers algorithm [2]Weiss S. & Indurkhya N, “Predictive Data Mining: A
Multilayer Perceptron algorithms gave an accuracy Practical guide”, Morgan Kauf-. Mann, 1998.
of 87.44%, when using whole data as a training data. [3]Technology Forecast: 1997 (1997), Price Waterhouse
Therefore it prove that MLP classifier prove to be World Technology Center, Menlo Park, CA.
very effective for prediction of slow learner. The
[4]William J. Frawley, Gregory Piatetsky-Shapiro, and
comparison graph of all used classifier is also given Christopher J. Matheus, “Knowledge Discovery in
in the result section. In our result after Databases: An Overview”, AI Magazine Volume 13
implementation of classifier INT_GRD, INT_ACC Number 3 (1992) (© AAAI)
and ATND are the most influencing attributes which
[5]BK Bhardwaj, S. Pal, “ Mining Educational Data to
affect the performance.
Analyze Students Performance”, (IJACSA), Vol. 2, No. 6,
In future, data may be collected for different 2011
school having different board of examination,
different area of operation and having different [6]M Bala, Dr. DB Ojha, “Study of applications of Data
Mining Techniques in Education”, (IJRST) 2012, Vol. No.
1, Issue No. IV, Jan-Mar ISSN: 2249-0604.

Imperial Journal of Interdisciplinary Research (IJIR) Page 746


Imperial Journal of Interdisciplinary Research (IJIR)
Vol-2, Issue-12, 2016
ISSN: 2454-1362, http://www.onlinejournal.in

[7]Ying Zhang, Samia Oussena, Tony Clark, Hyeonsook [16]Thulasi Kumarthulasi.kumar@uni.edu, University of
Kim, “Use Data Mining to improve student retention in Northern Iowa “Theoretical Basis for Data Mining
higher education – A CASE STUDY” Approach to Higher Education Research”.
[8]P. Cortez, and A. Silva (2008), “Using Data Mining To [17]N.V.Anand Kumar Research Scholar, Department of
Predict Secondary School Student Performance”, In Computer Science and engineering Anna university,
EUROSIS, A. Brito and J. Teixeira (Eds.), pp 5-12. Chennai G.V.Uma Assistant professor, Department of
Computer Science and Engineering Anna university,
[9]Galit.et.al, “Examining online learning processes based Chennai “Improving Academic Performance of Students
on log files analysis: a case study”. Research, Reflection, by Applying Data Mining Technique.
and Innovations in Integrating ICT in Education 2007.
[18]Aksenova, S. Sretlana., Zhang, Du and Lu, Meilin.
[10]M. Ramaswami and R. Bhaskaran (2010), “A CHAID “Enrolment prediction through Data Mining” IEEE
Based Performance Prediction Model in Educational Data Conference Publications (2006): 510-515
Mining”, International Journal of Computer Science Issues
Vol. 7, Issue 1, pp 10-18. [19]Baker, R.S.J.D., Data Mining for Education, in B.
McGraw, P. Peterson, and E. Baker (Eds.), international
[11]Nguyen Thai-Nghe, Andre Busche, and Lars Schmidt- Encyclopaedia of Education, Third Edition, Vol. 7, 2010,
Thieme (2009), “Improving Academic Performance 112-118
Prediction by Dealing with Class Imbalance”, Ninth
International Conference on Intelligent Systems Design [20]Cios, K.J., Swiniarski, W.R., Pedrycz, W. and Kurgan,
and Applications. A. L., Data Mining: A Knowledge Discovery Approach,
2007, Springer, New York
[12]V. Ramesh, P. Parkavi, K. Ramar (2013),”Predicting
student performance: A statistical and data mining [21]Kantardzic, Mehmed. Data Mining: Concepts, Models,
approach”, International journal of computer applications, Methods and Algorithm, Second Edition, John Wiley and
Volume 63- no. 8, pp 35-39. Sons, New Jersey, 2011
[13]Hideko Kitahama, “Data Mining through Cluster [22]Kardan, A. Ahmad., et al. “Prediction of student
Analysis Evaluation on Internationalization of Universities course selection in online higher education institutions” 65
in Japan”. (2013): 1-11.
[14]Bruce L. Golden R. H. Smith School of Business [23]Kumar, A.N.V. and Uma, G.V. “Improving academic
University of Maryland College Park, MD 20742 “An performance of students by applying data mining
Example of Visualization in Data Mining” techniques.” European Journal of Scientific Research, No.
4 (2009): 526-534
[15]Jing Luan, PhD Chief Planning and Research Officer,
Cabrillo College Founder, Knowledge Discovery [24]Kovacic, J. Zlatko, “Early prediction of student
Laboratories “Data Mining Applications in Higher success: Mining students’ enrolment data, in proceedings
Education”. of informing science and IT education Conferences
(InSITE) 2010, 647-665.

Imperial Journal of Interdisciplinary Research (IJIR) Page 747

You might also like