A Comparative Analysis of Different Machine Learning Algorithms Developed with Hyperparameter Optimization in the Prediction of Student Academic Success

Machine learning makes significant contributions in many areas of the applied sciences. One of these is the field of education, in the form of predicting students’ academic success and developing educational policies. In this study, two distance and kernel-based methods and eight tree-based and ensemble learning models were used to predict students’ academic success. The data set used in the study includes various variables, such as demographic information, academic information, course participation rates, and activity participation status, for 2392 students. Hyperparameter optimization was performed using genetic algorithm and grid search methods and model accuracy was tested with 10-fold cross-validation. In addition, the performances of all machine learning models were compared, using seventeen metric results for three cases, including results without hyperparameter optimization and determinations after hyperparameter optimization. Subsequent to the analyses performed, it was concluded that the SVR, GBM, and XGBoost methods have both high explanatory power and low error rates in regression problems requiring high accuracy, such as analyses aimed at predicting student success.