Prediction of On-time Student Graduation with Deep Learning Method

Abstract

Universities have an important role in providing quality education to their students so they can build a foundation for their future. However, a problem that often arises is that the process experienced will be different for each individual. Therefore, it is necessary to apply on-time graduation predictions for students with academic attributes in the hope that educational institutions can better understand student conditions and maximize on-time student graduation. In this study, a deep learning method was implemented to help predict on-time graduation for students at the Faculty of Computer Science, University of Brawijaya. Based on the test results and hyperparameter tuning using Optuna, the best hyperparameter configuration for the deep learning method consisted of number of layer combinations = 4; first-layer nodes = 118; first dropout = 0.3393; second-layer nodes = 83; second dropout = 0.0349; third-layer nodes = 88; third dropout = 0.0491; fourth-layer nodes = 65; fourth dropout = 0.4169; number of epochs = 244; learning rate = 0.0710; and optimizer = SGD. Thus, an accuracy rate of 86.61% was achieved for the two classes of the test data set, i.e., on-time graduation and not on-time graduation.