Evaluation of Random Forest and Support Vector Machine Models in Educational Data Mining

Abstract

The computer science field has witnessed the popularity of machine learning (ML) in discriminating low achieving and high-achieving students. However, various ML methods have different performances in predicting student performance. Therefore, the investigative analysis of their effectiveness in the discrimination of student based on their academic achievement would have been the major research concern these days. This study investigates the performance of the random forest (RF) and support vector machine (SVM) against their power in academic performance prediction of a student grade score (SGS). The analysis is performed based on the classification capability of the two algorithms using the Portuguese SGS dataset. Furthermore, the study also focused on the analysis of the impact of sigmoid and radial basis functions on the capability of the SVM for classifying SGS. We also presented a comparison among the various ML methods namely RF, and SVM, in identifying the student performance based on the SGS. Various demographic information (age, sex) and student assessment results (assignment, mid-term exam, and quiz) were used as the features in training. The result revealed that RF and SVM classifiers have the power to predict student performance. The SVM scored more accuracy than the RF. We obtained high accuracy (75.72%) using the linear kernel. The result implied that SGS can be predicted by using previous assessment results with the proposed SVM classifier.