Penerapan Metode Support Vector Machine dan Random Forest pada Klasifikasi Multikelas Minat Studi atau Karier Siswa Pasca Lulus SMA

Studi Kasus SMA Se-Kabupaten Kudus

Abstract

This study aims to compare the performance of the Support Vector Machine (SVM), Random Forest Feature Selection + SVM, and Random Forest Classifier methods in multi-class classification of post-high school students' study or career interests in Kudus Regency. The data used include academic and non-academic variables, such as subject grades, achievements, and parental support. The data processing process was carried out through a preprocessing stage that included handling missing values, categorical data transformation, and normalization. Model evaluation was carried out using the 10-Fold Cross Validation method with accuracy, precision, recall, and F1-score metrics. The results showed that the Random Forest Classifier model had the best performance with an accuracy of 47.63%, precision of 35.20%, recall of 27.52%, and F1-score of 30.9%. Meanwhile, the SVM and Random Forest Feature Selection + SVM models produced similar performance with an accuracy of 46.97% and an F1-score of 15.9%. Variable analysis showed that academic factors, especially Mathematics and Physics grades, were the most influential variables on student interest. However, the model's overall performance is still limited due to data imbalance and the lack of parameter optimization. This study shows that Random Forest is more effective in handling multiclass classification on educational data than SVM.