An empirical analysis of classification techniques for predicting academic performance

A multiclass classification refers to the classification of the instance into more than two classes. In real life many classification problem requires decisions among a set of contending classes. Multiclass classification and prediction is suitable for hand written digit recognition, hand written character recognition, speech recognition and body parts recognition etc. This paper compares five classification algorithms namely Decision Tree, Naïve Bayes, Naïve Bayes Tree, K-Nearest Neighbor and Bayesian Network algorithms for predicting students´ grade particularly for engineering students. This is a four class prediction problem. Student´s marks are classified into four classes A, B, C and F respectively. Initially complete data set is used to build the classifiers then Bootstrap method is used to improve the accuracy of the each classifier. Bootstrap method is a resample function available in WEKA tool kit. The excellent results of this function can be seen through IBK, Decision Tree and Bayes Net algorithm. However the overall results of all four algorithms are good but the results of individual classes for Naïve Bayes and NB Tree is not sufficient enough for the individual class prediction particularly for this study. This paper also presents a comparative study of the previous work related to student´s performance predictions.