Improving SVM Classification Performance on Unbalanced Student Graduation Time Data Using SMOTE

Abstract

Student graduation accuracy is one of the indicators of the success of higher education institutions in carrying out the teaching and learning process and as a component of higher education accreditation. So it is not surprising that building a system that can predict or classify students graduating on time or not on time is necessary for universities to monitor the exact number of students graduating on time using educational technology. Unfortunately, educational technology or machine learning with data mining approaches is less accurate in classifying classes with unbalanced data. Therefore, this research purpose is to build a machine learning system that can improve classification performance on unbalanced class data between students who graduate on time and graduate late. This study applies the Synthetic Minority Oversampling Technique (SMOTE) method to improve the classifying performance of the Support Vector Machine (SVM) data mining method. The results of the study concluded that using the SMOTE method increased the accuracy, precision, and sensitivity of the SVM method in classifying class data of unbalanced student graduation times. The SVM performance score rises by 3% for classification accuracy, 8% for classification precision, and 25% for classification sensitivity.