Performance Analysis of Machine Learning Algorithms in Prediction of Student Academic Performance
Michael Donkor Adane *
Department of Mathematics and ICT, Akatsi College of Education, P. O. Box PMB, Akatsi, Ghana.
Joshua Kwabla Deku
Department of Mathematics and ICT, E.P. College of Education, P. O. Box AM 12, Amedzofe, Ghana.
Emmanuel Kwaku Asare
Department of Mathematics and ICT, Presbyterian Women’s College of Education, Aburi, Ghana.
*Author to whom correspondence should be addressed.
Abstract
The advancement in technology has contributed largely to the application of data mining in education in recent times. However, selecting appropriate algorithm(s) to “mine” knowledge about educational data presents a difficult challenge to researchers and analyst. This paper contributes to the use of classification algorithms in academic performance prediction. The predictive ability of four popular algorithms; C4.5 Decision tree (CDT), Multilayer Perceptron (MLP), Naïve Bayes (NB) and Random Forest (RF) algorithms were compared. The models were built using student dataset from selected private senior high schools in Ghana. The comparative analysis of the algorithms was made based on their Accuracy, Recall, Specificity, F-Measure and Running time. On all the training and test ratios; 80:20, 70:30 and 10-fold cross validation, the results indicated that all the algorithms performed well in the classification. However, the Naïve Bayes algorithm performed significantly better than the MLP and CDT on some ratios. The running time of the NB, CDT and RF were the quickest while MLP took the longest time.
Keywords: Data Mining, algorithms, machine learning, classification, prediction, student performance, multilayer perceptron algorithm, naïve bayes algorithm, C4.5 decision tree