Comparative Analysis of Random Forest Regressor, Support Vector Regression, and MLP Regressors for Student Graduation Time Prediction

Abstract

Indonesian bachelor degree students are considered graduated in time when the student graduated in 4 year or less with lower graduated times are preferrable. As such, accurate graduation time prediction is critical for institutional resource planning and academic intervention strategies as it impacts the public perceptions of the institutions and its accreditation. Using regression method, this research will test three machine learning model, namely Random Forest Regressor, Support Vector Regression, and Multilayer Perceptron Regressor to predict the total study in years. Compared to the previous approaches, which relies on grade point average (GPA), this approach uses more granular academic study report for each semester and every courses grades. The student data are collected from Department of Informatics, Soegijapranata Catholic University which enrolled between 2018 and 2021. To evaluate the model, each model processed the semester-by-semester student data for every students as the input features and evaluated using several regression scoring method such as R-squared and MSE to picture the dataset generalization. The goal of this research is to lay foundation for further research on early warning systems, especially for department heads to give ability to identify students and to formulate intervention strategies for students with risks of delayed graduation.