PERBANDINGAN ALGORITMA K-NEAREST NEIGHBOR DAN NAÏVE BAYES DALAM MEMPREDIKSI WAKTU KELULUSAN MAHASISWA

Abstract

Students are an important aspect for higher education institutions, especially regarding the time of student graduation. Therefore, it is critical to know the prediction of the time length for completing studies. This study proposes creating a prediction system for student graduation rates; hence it could be a preventive measure for students to improve their learning process. This research used machine learning techniques to compare the K-Nearest Neighbor (KNN) and Naïve Bayes algorithms. The experiment aimed to determine the best model, such as the amount of data collection, the number of classification classes, and the handling of imbalanced classes. Based on all experiments, the KNN method achieved higher results than the Naïve Bayes method. Applying the SMOTE oversampling technique significantly increased the difference in evaluation scores (precision, recall, F1 score, and accuracy) between 12% and 41% in the Naïve Bayes and KNN methods. The results of the 4-class prediction model using the KNN method with SMOTE get a precision value of 79%, a recall value of 78%, an F1 score of 78%, and an accuracy of 78%. In comparison, the prediction results for eight classes using the KNN method with SMOTE get precision, recall, F1 Score, and accuracy values of 93%.