Mukt Shabd Journal ISSN NO : 2347-3150

A Feature Selection technique based Approach for predicting Student

Academic Performance

Dr.T. Murali Mohan1, T V Satya Sheela2, Archna.Gelli3

1
Professor of CSE and Principal, Swarnandhra Institute of Engineering and Technology,
Narsapuram, West Godavari, AP
2
Assistant Professor, Department of CSE, Swarnandhra Institute of Engineering and
Technology, Narsapuram, West Godavari, AP
3
Assistant Professor, Department of CSE, Swarnandhra college of Engineering and
Technology, Narsapuram, West Godavari, AP
1
[email protected], [email protected], [email protected]

Abstract: Education is a primary need for every citizen in a nation. In every year, several students are
successfully completing their education and doing various jobs. However, many students are failed in their
career because of several factors. There is a need of one research filed to analyse which factors are more
influencing the failure of a student. Educational data mining is one such research which is used to analyse the
student dataset and extracting the hidden patterns in the student dataset. The data mining techniques are
implemented in educational data mining to extract hidden information. The existing student datasets are
described with several features or attributes. All features that are specified in the dataset are not useful for
analysing the student performance. To identify the most relevant and important features, feature selection
algorithms are used. In this work, a feature selection algorithms based approach is proposed for predicting the
student academic performance. In this approach, a feature selection algorithm is used to identify the best
informative and relevant features for improving student performance. The UCI student dataset is used in this
work for evaluating the performance of students. Different machine learning algorithms are used in this work for
evaluating the proposed approach. The Random Forest classifier attained best accuracy for predicting the
student academic performance.

Key Words: Educational data Mining, Data Mining, Machine Learning Algorithms, UCI dataset, Feature
Selection Algorithms

1. Introduction

Education is a primary need for every human being in the society to develop ethical values, enhancing the
awareness about surroundings and increasing the employment skills in the humans. Several factors like student’s
attendance, attitude, listening skills, school reputation etc., are influencing the student rate of success or failure
in their education. The administrators and academicians are continuously struggling a lot to reduce the failures
of a student in throughout the world. They are following various methods such as giving assignments,
conducting additional tests, remedialclasses and counselling the students to enhance the student academic
performance. However, the number of dropout students, poor performances of students and failure rates of
students are increasing in every year which disturbs the parents and academicians. The researchers come into the
picture and identifying different type of features such as individual attributes, demographic, economic and social
features that influence the performance of a student. More specifically, some set of features like internetusage
hours, alcohol consumption, parental support,parents’ occupation,socialbackground,health and student
attendance are affecting more on performance of a students
Researchers are developing a research field called as Educational Data Mining (EDM) to handle the
problems of enhancing the Student Academic Performance (SAP). EDM is one research field which is used for
enhancing the student performance by implementing machine learning and deep learning techniques. The
researchers applied data mining techniques on the student dataset to extract the hidden patterns that are
interesting and more useful to the academicians to take strategic decisions for improving the performance of a
student. The EDM techniques are used for various purposes such as prediction of student performance, E-
studentmodelling, detection of learners behaviours, curriculum design, developing recommendation system for

Volume X, Issue IX, SEPTEMBER/2021 Page No : 892

Mukt Shabd Journal ISSN NO : 2347-3150

learners, student data visualization and in various other fields. EDM is a combination of various data mining
techniques that are mainly used to enhance and analyse the methods that are used in the process of teaching and
learning
In data mining techniques based analysis of approaches, the major problem faced with huge number of
features in the dataset which also called as high dimensionality [1]. The datasets contain large number of
features, but all features are not necessarily improving the performance of the proposed approaches. In EDM
techniques also, the student dataset contain more number of features, but all features like roll-no, name etc., are
not useful for analysis. Many researchers worked on this problem and developed a concept of feature selection
algorithm to reduce the number of features for analysis. The feature selection algorithms identify important and
relevant features from a set of features. Education is a compulsory for everyone in the nation and the process of
education is influenced by various factors like nation’s economic prosperity, education policies etc. [2]. The
identification of weak students at starting time of any degree programme was more helpful for the academicians
for taking any necessary plans and action to enhance the academic performance of a student [3].
Several researchers proposed various machine learning algorithms for analysing student’s knowledge in a
better way by extracting different types of knowledge that are hidden in the dataset. Most of the algorithms
proved to reduce the student failures count by estimating the performance of a student accurately [4]. The
Feature selection algorithms usage in data mining techniques profoundly influences the performance of student
academic performance prediction. The feature selection algorithms are used for various purposes such as reduce
the count of features, reducing the training time, decreasing the over fitting problem by enhancing generalization
and avoiding the curse of dimensionality. In data processing and analysis field, the datasets contain various
attributes or features that are used to assess the applicability and usability of the data [5]. Another challenging
problem faced by classification algorithms are the problem of data imbalance [6].
The student datasets contain huge number of features to describe a student. Several features or factors are
affecting the academic performance of a student, but it is not easy to recognize which set of features are
influencing more for enhancing student performance. To handle this problem, feature selection algorithms are
used to identify which features are more important and relevant for affecting student performance. In this work,
a feature selection method based approach proposed for student academic performance prediction. in this
approach, we used a feature selection algorithm to identify the relevant features for enhancing student
performance. The UCI student dataset is used in this experiment to evaluate the performance of proposed
approach. Different machine learning algorithms such as K-Nearest Neighbour (KNN), Logistic Regression
(LR), Support Vector Machine (SVM) and Random Forest (RF) are used in this experiment for generating
classification model.
This paper is organized in eight sections. The section 2 describes the related works proposed by the
researchers. The information about dataset is explained in section 3. Machine Learning algorithms are described
in section 4. The evaluation measures are discussed in section 5. The proposed approach for enhancing the
student academic performance is described in section 6. The experimental results are discussed in section 7. The
section 7 lists the conclusions and future enhancements to this work.

2. Related Work

The data mining techniques are used by the researchers for predicting the student academic performance.
Concepción Burgos et al., developed [7] data mining techniques to the student history dataset for predicting the
student dropouts in a course. They implementedLR model for filtering irrelevant features and for data
classification. The proposed technique efficiency was tested on different distance learning courses data of 100
studentsthose are registered in Madrid Open University. The researchers identified that when compared with
previous academic years the dropouts are reduced by 14% with the proposed technique.
GetanehBerieTarekegn et al., implemented [8] techniques of data mining for predicting the student performance
in campus placements of various departments. The experiment conducted with a student dataset that contains the
results of entrance examination. Three machine learning techniques such as J48, RF and Naïve Bayes (NB) were
applied on the dataset to generate the classification model for prediction. The researchers identified that the RF
classifier performance was good than other two techniques for predicting whether students got placement in the
companies or not. Radwan and Cataltepeaimplemented [9] two noise reduction methods for solving the problem
of imbalance in classesof dataset. The combination of over-sampling technique of SMOTE and threshold
method were used in the proposed method for balancing the training dataset and to choose the best boundary
among classes. The proposed method is validated by usingUCI Portuguese dataset. it was identified that the
developed method shows good performance in the noise reduction process. The authors also identified that the
parameter selection methods are more helpful for enhancing the student performance prediction.
Eduardo Fernandes et al., developed [10] a method and implemented this method on the public school dataset
that was collected from of the Federal District of Brazil. The proposed method was applied for analysing the
data and method efficiency was tested on two datasets. The student performance was predicted by using

Volume X, Issue IX, SEPTEMBER/2021 Page No : 893

Mukt Shabd Journal ISSN NO : 2347-3150

Gradient Boosting Machine (GBM) classifier. The authors identified from results of experiment that the
residence of a student and school are two factors that are affecting the student academic performance more. The
authors identified that the requirement of different factors for improving the student performance. RaheelaAsif,
et al., considered [11] two tasks such as predicting the performance of a student at the end of academic
programme of four-year and possible progressions of students. The dataset contains two classes such high and
low achieving students. They used decision tree for feature selection and applied k-means algorithm for
classification. The proposed method is tested on the dataset for predicting the performance of a student. They
observed that the experimental results of proposed method show good performance.
Ahmad [12] used the student database of July 2006 to July 2014 which containsthe student academic records,
demographics information, and information related tofamily background. They experimented with Naïve Bayes,
decision tree, and rulebasedclassification algorithms to predict the student performance. It was found thatthe
rule-based classifier obtained the highest accuracy of 71.3% for predicting studentacademic performance among
three classifiers. Li et al.,solved [13] the class noise and imbalance problems in the dataset by proposing cost
function based randomized learning algorithm. The randomized learning algorithm developed by using weight
least square problem. The performance of proposed method is compared with existing standard methods for
predicting the performance of student. The experimental results show that the proposed method solved the
problems of label errors and data imbalance problems.
Koutina [14] used machine learning techniques to predict the performance of postgraduatestudents. They used
the postgraduate students database of Ionian UniversityInformatics. They experimented with six classifiers such
as J48 decision tree classifier,K-nearest neighbor classifiers (1-NN, 3-NN, 5-NN), Naïve Bayes, JRIP,
randomforest, and support vector machines. It was observed that Naïve Bayes and 1-NNclassifiers along with
feature selection algorithm obtained good accuracy results forpredicting student performance.
Son and Fujita developed [15] an adaptive fuzzy technique to enhance the prediction accuracy of student
performance. The proposed technique process both local and global learning. In local learning, the settings of
parameters in fuzzy technique were done by using hybrid method. In global learning, the random subsets were
selected for training. In the parameter learning, the hybrid method is the combination of Particle Swarm
Optimization and Gradient descent methods. The proposed method performance is validated by using different
datasets of UCI and compared with various standard methods. The performance of learning rate and
classification are needed to be enhanced for attaining best efficiency from the model.
In [16], the authors experimented with Decision Trees (DTs) and Neural Networks (NNs) for predicting the
achievements of students in academics. They prepared the 22 features student dataset that contains the
information of 1600 students those are joined in Thailand universities among 2001 to 2011 year. The features in
thedatabase are student ID, age,gender, educational background, status of student (single, married, and
divorced),qualification,continent, mother occupation,father occupation, on-campus residence, nativity language,
scholarship status, pre-university English,department, numberof credits, grade point average, highest cumulative
grade point average of subject, cumulative grade point average, extra study hour,extracurricular subject,number
of activity hours and number of work hours. It was identified that the DT classifier attained best accuracy than
NNs. They also identified that the students which are having good communication skills in English shows best
academic performance.
Khan [17] used educational datamining techniques for prediction of scholarship ofstudents. They analyzed
student data with ID3 and J48 decision tree algorithms forpredicting winning chances of scholarship by
converting decision tree into IF-THENrules. The features—the rank or position in a class, grades in a semester,
minimumand maximum number of credit hours taken, achievements and extracurricular activities—were used to
predict the scholarship winning chances. It was observed thatID3 achieved good accuracy when compared with
J48 even though J48 generates asmaller tree and faster in classification than ID3.
Agrawal [18] experimented with four classification algorithms such as randomforest, J48, Naïve Bayes, and
multilayer perception and seven attribute evaluationtechniques such as Filtered Attribute Eval, Chi-Squared
Attribute Eval, Info GainAttribute Eval, Gain Ratio Attribute Eval, OneR Attribute Eval, ReliefAttributeEval,
and Symmetric Uncert Attribute Eval.It was observed that the J48 algorithm with no attribute evaluator method
performancewas good when compared the performance of J48 alone. It was also observedthat the random forest
algorithm perform well when combined with Relief AttributeEval, Naïve Bayes algorithm performance is good
when combined with FilteredAttribute Eval, Symmetric Uncert Attribute Eval, and finally the combination
ofmultilayer perception algorithm and Chi-Squared Attribute Eval obtained good performance
for prediction of student performance in exams.
Siri [19] predicts the dropouts of students at University of Genoa in the academicyear 2008–09 using artificial
neural networks. They used the database of 810 studentsregistered for healthcare professions degree and
collected the information throughtelephone conversation with students who were not registered for successive
years,administrative information relevant to student careers and statistical data collectedthrough ad hoc survey.

Volume X, Issue IX, SEPTEMBER/2021 Page No : 894

Mukt Shabd Journal ISSN NO : 2347-3150

3. Dataset Characteristics

The UCI dataset is collected from two Portuguese schools of secondary education. The dataset was obtained
through school questionnaires and reports. The two datasets contains student performance in two different
subjects such as Portuguese language (por) and Mathematics (mat). The table 1 presents the features of UCI
student dataset. The two datasets of UCI (Mathematics) and UCI (Portuguese) contains 33 features.

Table 1. Characteristics of UCI student dataset

Volume X, Issue IX, SEPTEMBER/2021 Page No : 895

Mukt Shabd Journal ISSN NO : 2347-3150

4. Machine Learning algorithms

In a document classification problem, the goal is to classify the class of an unseen document among the set of
previously seen classes in the training set. The vector representation of the document is used to train the
machine learning algorithm. Any given document d either seen or unseen is represented as a feature vector. The
goal of a closed-class classification task is, given an unseen document d, identify the class of this document
among the set of authors based on the knowledge gained through the training set. In general, Classifier is also
called as learning method and mainly it has three parts. As the first step a model is created with the help of
algorithm on training data. This process is termed as modeling, training model involves in training on training
data and lastly testing part. In the classification process in order to test the classifier a special testing method
called K-fold cross-validation testing method was used. This method loops throw K iterations. In Every
iteration, to test a single unit datum a random selection procedure is used and the left over K-1 datum’s were
utilized for training. In this work four machine learning algorithms such as KNN, LR, SVM and RF are used in
the experimentation.

Volume X, Issue IX, SEPTEMBER/2021 Page No : 896

Mukt Shabd Journal ISSN NO : 2347-3150

5. Evaluation Algorithms

The machine learning algorithms are presenting the performance of proposed approaches in the form of
evaluation measures like Precision, recall, F1-score, accuracy etc. To describe the evaluation measures,
contingency table is required which is displayed in Table 2.

Table 2. Contingency table

Original labels of documents

Class Ci
Original YES Original NO

Predicted TP FP

Predicted by YES (True Positives) (False Positives)

the system Predicted FN TN

NO (False negatives) (True Negatives)

Where TP is number of YES label samples are predicted as YES label, TN is number of NO label samples are
predicted as NO label, FN is number of NO label samples are predicted YES label and FP is number of YES
label samples are predicted as NO label.
Precision is the ratio between the count of positive class documents predicted as positive class and the count of
documents predicted as positive from documents of positive and negative classes. Precision is represented in
Equation (1).
TP
Pr ecision  (1)
TP  FP

The recall is the ratio among the number of documents correctly predicted as positive class and the count of
samples in positive class. Recall is denoted in equation (2).

TP
Re call  (2)
TP  FN

F1-Score is the harmonic mean of recall and precision. F1-score is computed by using equation (3)

2  Pr ecision  Re call
F1  (3)
Pr ecision  Re call

Accuracy is sum of count of positive class documents predicted as positive class and count of negative class
documents predicted as negative class divided by total documents count in the training dataset. The accuracy is
defined in equation (4).
Number of test samples correctly predicted (TP+TN)
Accuracy  (4)
Total number of test samples (TP  FP  FN  TN )

In this work, accuracy measure is used to display the results of student academic performance prediction.

6. Feature Selection Algorithm based Approach for Student Academic Performance Prediction

Fig. 1 shows the proposed approach for student academic performance prediction. In this proposed approach,
extract all features from the UCI student dataset. All features are given to feature selection algorithms and these
algorithms identify the important features from all feature set. The student records in the dataset are represented
with reduced features as vectors. The student vectors are passed to machine learning algorithms. The machine
learning algorithms develop a classification model and this model is used for predicting the academic

Volume X, Issue IX, SEPTEMBER/2021 Page No : 897

Mukt Shabd Journal ISSN NO : 2347-3150

performance of a new student vector. The next section explain the feature selection algorithm that was used in
this proposed approach.

Fig. 1. Proposed Approach

6.1 Feature Selection Algorithm

In general, most of the features are extracted to analyse the datasets are a combination of incomplete, noisy,
inappropriate, and outlier. Selection of best features plays a very important character to get accurate, appropriate
result and in improving the processing time. The researchers are concentrated to investigate on feature selection
to address this problem. Several researchers proposed a lot of statistical approaches and feature selection
algorithms to choose the feasible and appropriate features within the corpus. In recent times, selection of best
features has become a dynamic research domain in text classification and information retrieval. The primary
goal of feature selection algorithms are choosing the more informative features within the actual feature list, also
eliminating those features which are having no predictive information [20, 21]. Feature selection algorithms are
categorized into Unsupervised Feature Selection Algorithms (UFSA) [22, 23], Supervised Feature Selection

Volume X, Issue IX, SEPTEMBER/2021 Page No : 898

Mukt Shabd Journal ISSN NO : 2347-3150

Algorithms (SFSA) [24] and Semi-Supervised Feature Selection Algorithms (SSFSA) [25]. In UFSA, the
training instances are not having class labels. In SFSA, the class labels are associated to all instances. In SSFSA,
some instances of total training instances have class labels [26].
In this work, we used a Correlation based Feature Subset Selection (CFSS) [27] algorithm to identify the best
informative terms to distinguish the student records. CFSS algorithm evaluates on the subsets of attributes rather
than individual attributes. This algorithm allocates high ranks to the subsets of features which are having high
correlation with the interested class and strong correlation among the features in a class. The Equation (5) is
used to compute the correlation among the subset of attributes and class.

k r cf
M e r its 
k  k ( k  1) r f f (6)

Where, Merits is the “merit” of a feature subset S which contains k features, rcf is the average correlation among
class and feature subset and rff is the average correlation among individual features. This algorithm identified a
subset of 14 features as best informative features.

7. Experimental Results

In this work, the experiment conducted for predicting the student academic performance. The UCI dataset is
used for experimentation purpose. This dataset contains 33 features. The proposed approach identify 14 features
are more relevant features and these features are used for representing the student vectors. The accuracies of
different machine learning algorithms are displayed in Table 3.

Table 3. The Accuracies of Student Academic Performance Prediction

Machine
Accuracy
Learning
(14 Features)
Algorithms
KNN 80.67

LR 83.79

SVM 85.53

RF 87.34

In Table 3, The RF classification algorithm attained best accuracy of 87.34% for Student Academic performance
prediction than other classification algorithms. The SVM, LR and KNN attained accuracies of 85.53%, 83.79%
and 80.67% respectively for predicting student academic perfromance.

8. Conclusions and Future Scope

The aim of any nation is maintaining all educated people in all places. The education improves the life style of
humans. Monitoring of students success and failures is one important task in any college or university. The
colleges maintain student databases for examining the characteristics of students. EDU is one field which uses
data mining techniques to extract the useful and hidden information from the student datasets. This information
was used by academicians to take decisions about students. The data mining algorithms used the features of
student information to build a classification model. All features in the student dataset are not useful for analysis.
To identify the best informative and relevant features, feature selection algorithms were used. In this work a
feature selection algorithm based approach was proposed for student academic performance prediction. The UCI
dataset is used for experimentation. Different machine learning algorithms such as KNN, LR, SVM and RF are
used to evaluate the performance of proposed approach.The RF shows best accuracy for student academic
performance prediction than other algorithms.
In future work, we are planning to implement deep learning techniques for predicting the student academic
performance prediction.

Volume X, Issue IX, SEPTEMBER/2021 Page No : 899

Mukt Shabd Journal ISSN NO : 2347-3150

References

1. Wang XD, Chen RC, Yan F, et al. Fast adaptive K-means subspace clustering for high‑dimensional
data. IEEE Access. 2019;7:42639–51.
2. J. Xu, K. H. Moon, and M. V. D. Schaar, “A machine learning approach for tracking and predicting
student performance in degree programs”, IEEE Journal of Selected Topics in Signal Processing,
Vol.11, No.5, pp.742-753, 2017.
3. Pandey, M. and Taruna, S., 2016. Towards the integration of multiple classifier pertaining to the
Student's performance prediction. Perspectives in Science, 8, pp.364-366.
4. ASSELMAN, A., KHALDI, M. and AAMMOU, S., 2020. Evaluating the impact of prior required
scaffolding items on the improvement of student performance prediction. Education and Information
Technologies, pp.1-23.
5. Raghunadha Reddy T, Vishnu Vardhan B, Vijayapal Reddy P, “A Survey on Author Profiling
Techniques”, International Journal of Applied Engineering Research, March 2016, Volume-11, Issue-5,
pp. 3092-3102.
6. Chen RC. Using deep learning to predict user rating on imbalance classification data. IAENG Int J
Comput Sci. 2019;46:109–17.
7. C. Burgos, M. L. Campanario, D. de la Peña, J. A. Lara, D. Lizcano, and M. A. Martínez. (2018). Data
mining for modeling students’ performance: A tutoring action plan to prevent academic dropout.
Computers & Electrical Engineering, 66, pp. 541-556.
8. GetanehBerieTarekegn, VudaSreenivasarao, “ Application of Data Mining Techniques to Predict
Students Placement into Departments ”, in International Journal of Research Studies in Computer
Science and Engineering (IJRSCSE), Volume 3, Issue 2, 2016, PP 10–14.
9. Radwan, A.M. and Cataltepe, Z., 2017. Improving performance prediction on education data with noise
and class imbalance. Intelligent Automation & Soft Computing, pp.1-8.
10. E. Fernandes, M. Holanda, M. Victorino, V. Borges, R. Carvalho, and G. Van Erven. (2019).
Eduational data mining: Predictive analysis of academic performance of public school students in the
capital of Brazil. Journal of Business Research, 94, pp. 335-343.
11. R. Asif, A. Merceron, S. A. Ali, and N. G. Haider. (2017). Analyzing undergraduate students'
performance using educational data mining. Computers & Education, 113, pp. 177-194.
12. Fadhilah Ahmad*, NurHafieza Ismail and Azwa Abdul Aziz, “The Prediction of Students’ Academic
Performance Using Classification Data Mining Techniques”, Applied Mathematical Sciences, Vol. 9,
2015, no. 129, 6415–6426.
13. Li, M., Huang, C., Wang, D., Hu, Q., Zhu, J. and Tang, Y., 2019. Improved randomized learning
algorithms for imbalanced and noisy educational data classification. Computing, 101(6), pp.571-585.
14. Maria Koutina, Katia LidaKermanidis, “Predicting Postgraduate Students’ PerformanceUsing Machine
Learning Techniques”, Artificial Intelligence Applications and Innovations, pp159–168.
15. Fujita, H., 2019. Neural-fuzzy with representative sets for prediction of student performance. Applied
Intelligence, 49(1), pp.172-187.
16. PimpaCheewaprakobkit, “ Predicting Student Academic Achievement by Using the Decision Tree and
Neural Network Techniques”, in catalyst, Volume 12, No. 2, 2015, pp 34–43.
17. IrfanAjmal Khan and Jin Tak Choi, “An Application of Educational Data Mining (EDM) Technique
for Scholarship Prediction”, in International Journal of Software Engineering and Its Applications, Vol.
8, No. 12 (2014), pp. 31–42.
18. RichaShambhulalAgrawal, 2Mitula H. Pandya, “Data MiningWith Neural Networks to Predict
Students Academic Achievements”, in International Journal of Computer Science And Technology,
Vol. 7, Issue 2, April - June 2016, PP. 100–103.
19. Anna Siri, “Predicting Students’ Dropout at University Using Artificial Neural Networks”, inITALIAN
JOURNAL OF SOCIOLOGY OF EDUCATION, 7 (2), 2015, PP 224–247.
20. T. Raghunadha Reddy, P. Vijayapal Reddy, T Murali Mohan, RajuDara, “An Approach for Suggestion
Mining based on Deep Learning Techniques”, International Conference on Computer Vision, High
Performance Computing, Smart Devices and Networks(CHSN-2020), 28-29 December, 2020, JNTUK,
Kakinada, Andhra Pradesh.

Volume X, Issue IX, SEPTEMBER/2021 Page No : 900

Mukt Shabd Journal ISSN NO : 2347-3150

21. P Buddha Reddy, Dr. T Murali Mohan, Dr. P Vamsi Krishna Raja and Dr. T Raghunadha Reddy, “A
Novel Approach for Authorship Verification”, SPRINGER 3rd International Conference on Data
Engineering and Communication Technology (ICDECT), Stanley College of Engineering and
Technology for Women, Abids, Hyderabad, Telangana, India, 15 – 16 March, 2019.
22. Raghunadha Reddy T, Vishnu Vardhan B, Vijayapal Reddy P, “Profile specific Document Weighted
approach using a New Term Weighting Measure for Author Profiling ”, International Journal of
Intelligent Engineering and Systems, 9 (4), pp. 136-146, Nov 2016.
23. Raghunadha Reddy T, Vishnu Vardhan B, Vijayapal Reddy P, “A Document Weighted Approach for
Gender and Age Prediction”, International Journal of Engineering -Transactions B: Applications,
Volume 30, Number 5, pp. 647-653, May 2017.
24. Liu H and Motoda H , “Computational Methods of , Feature Selection”, Chapman & Hall/CRC Taylor
& Francis Group 6000 Broken Sound Parkway NW, Suite 300 Boca Raton, FL 33487- 2742 © 2008 by
Taylor & Francis Group, LLC Chapman & Hall/CRC is an imprint of Taylor & Francis Group, an
Informa business. http://www.crcpress.com, 2008.
25. Bijanzadeh E, Emam Y and Ebrahimie E., “Determining the most important features contributing to
wheat grain yield using supervised feature selection model”, Australian Journal of Crop Science, 2010,
Volume 4, Issue 6, pp. 402-407.
26. Srikanth Reddy G, Murali Mohan T, Raghunadha Reddy T, “Author Profiling Approach for Location
Prediction”, first international conference on Artificial Intelligence and Cognitive computing conducted
by MLR Institute of Technology, Dundigal, Hyderabad, 02-03 February, 2018.
27. HALL, M. A. & HOLMES, G. 2003. Benchmarking attribute selection techniques for discrete class
data mining. Knowledge and Data Engineering, IEEE Transactions on, 15, 1437-1447.

Volume X, Issue IX, SEPTEMBER/2021 Page No : 901