Accuracy Enhancement of Prediction Method using SMOTE for Early Prediction Student's Graduation in XYZ University

Abstract

According to the Minister of Education and Culture of the Republic of Indonesia's regulations from 2014, one of the essential elements in implementing higher education is the student's study duration. Higher education institutions will use early graduation prediction as a guide when developing policy. According to XYZ University data, the student study period is Grade Point Average (GPA), Gender, and Age are all aspects to consider. Using a dataset of 8491 data, the Prediction of Early Graduation of Students based on XYZ University data was examined by this study, particularly in the information systems and informatics study program. The aim is to find significant features and compare three prediction models: Artificial Neural Networks (ANN), K-Nearest Neighbor (K-NN) method, and Support Vector Machines (SVM). The Challenge in the development of a prediction model is imbalanced data. The Synthetic Minority Oversampling Technique (SMOTE) handles the class imbalance problem. Next, the machine learning models are trained and then compared. Prediction results increase. The best test accuracy value is on ANN with a data Imbalance of 62.5% to 70.5% after using SMOTE, compared to the accuracy test on the K-NN method with SMOTE 69.3%, while the SVM method increased to 69.8%. The most significant increase in recall value to 71.3% occurred in the ANN.