Analysis of Students' Academic Performance Using Traditional And Machine Learning Classifiers

Abstract

Unlike traditional statistical techniques of analyzing data, machine learning algorithms have made it easier to feed computer with (partitioned) dataset then slightly program it in order to obtain a model with the most precise classification. This study juxtaposed the classification and predictive performance of traditional statistical classifiers as compared with machine learning algorithms. Simultaneously, the study aim to classify and predict students' academic performance in Federal polytechnic Ede, Osun state. The considered classifiers include Logistic regression, Decision Tree, support vector machine and Naive Bayes. Students’ academic performance was classified as a factor of other variables which include; Age at admission, Years before admission, Program type, High school type, 0'level, and Gender. These classifications were subjected to varying train-test ratio ranging from 90-10 to 50-50. It was obtained that some of our classifiers performs below 50% accuracy, however decision tree algorithm yielded the most precise classification has it outperforms both Naïve Bayes and logistic regression largely, but support vector machine slightly. It is highly evident that all classifiers yielded maximum individual accuracies in 90-10 train-test ratio. Thus, we conclude that the decision tree algorithm under 90% training set and 10% testing set is appropriate for future prediction of students’ academic performance and it is recommended that it should be adopted for similar studies.