Pemodelan Support Vector Machine Data Tidak Seimbang Keberhasilan Studi Mahasiswa Magister IPB

Abstract

Bogor Agricultural University Postgraduate School (SPs-IPB) can maintain its reputation by applying a more selective admissions system. This research predicts the success of student using Support Vector Machine (SVM) modeling by considering the characteristics and educational background of the students. But there is an imbalance of data class. SVM modeling on unbalanced data produces poor performance with a sensitivity value of 0.00%. Unbalanced data handling using Sythetic Minority Oversampling Technique (SMOTE) succeeded in improving SVM classification performance in classifying unsuccessful students. Based on accuracy, sensitivity, and specificity with the default cut off, the exact type of SVM to model student success is SVM RBF. When using the optimum cut-off value from each type of SVM, the sensitivity value can be improved again. SVM RBF still gives the best result when using cut off 0.6. The final model that will be used to predict the success of the SPs-IPB student is obtained from SVM RBF modeling with cut off 0.6 using the entire data that has been through the SMOTE stage.