To ameliorate classification accuracy using ensemble vote approach and base classifiers

Abstract

Ensemble methods in the realm of educational data mining are embryonic trends deployed with the intent, to ameliorate the classification accuracy of a classifier while predicting the performance of students. In this study, primarily miscellaneous mining classifiers were applied on our real academic dataset to foretell the performance of students and later on, ensemble vote (4) method was employed wherein the hybridization of predicted output was carried out, with majority vote as consolidation rule. The various classification techniques used in our study vis-à-vis Naive Bayes, k nearest neighbours, conjuctive rules and Hoeffding tree. The empirical results attained corroborate that there is a paramount significance in performance after application of ensemble vote (4) method. Furthermore, a novel attempt was made wherein, the pedagogical dataset was subjected to filtering process, viz., synthetic minority oversampling technique (SMOTE), with the intention to verify whether there is a considerable improvement in the output or not. The findings by and large have clearly confirmed that after application of SMOTE, the classifier achieved high accuracy of 98.30% in predicting the resultant class of students. Therefore, it calls upon the researchers to widen the canvas of literature by utilising the similar methods to unearth the different patterns hidden in datasets.