Meningkatkan kinerja SVM: Dampak berbagai teknik seleksi fitur pada akurasi prediksi

Abstrak

In Higher Education accreditation, student graduation plays an important role as one of the assessment criteria. Graduation prediction is the main focus of helping institutions assess a student to graduate on time. This study takes historical data from students who have graduated, which is taken through a questionnaire from the Department of Information Systems and Informatics Engineering at Semarang University students. The feature selection selects the most relevant attributes in graduation prediction. The results of this selection are tested using the Support Vector Machine (SVM) Algorithm. The main objective of this study is to evaluate the impact of feature selection on graduation prediction. The results show that SVM with feature selection using weight by relief achieves an accuracy of 82%, a precision of 83.42%, and a recall of 80.83%. In contrast, SVM without weight by relief shows an accuracy of 69.23%, a precision of 70.83%, and a recall of 67.86%. The use of feature selection successfully reduces features from 27 to four of the most influential features..