Perbandingan Kinerja Random Forest, Decision Tree, dan Naive Bayes Menggunakan Metode 10-Fold Cross Validation untuk Prediksi Status Akademik Mahasiswa

Abstract

Predicting students' academic status is an important effort that higher education institutions can undertake to identify students who are at risk of experiencing academic decline, delayed graduation, or dropout. Early identification enables institutions to provide appropriate academic and non-academic interventions, thereby improving student retention and graduation rates. This study aims to compare the performance of the Random Forest, Decision Tree, and Naive Bayes algorithms in predicting students' academic status using the Predict Students Dropout and Academic Success dataset obtained from the UCI Machine Learning Repository. The dataset consists of 4,424 student records with 36 predictor attributes and three target classes: Dropout, Enrolled, and Graduate. The research methodology includes data exploration, feature selection, model development using Orange Data Mining, and model evaluation through the 10-Fold Cross Validation method. Model performance was assessed using Accuracy, Precision, Recall, F1-Score, Area Under the Curve (AUC), Matthews Correlation Coefficient (MCC), Confusion Matrix, and ROC Curve. The experimental results indicate that the Random Forest algorithm achieved the best performance, with an Accuracy of 77.4%, Precision of 0.760, Recall of 0.774, F1-Score of 0.759, AUC of 0.897, and MCC of 0.624, outperforming both Decision Tree and Naive Bayes. This study contributes a comparative evaluation of three widely used classification algorithms on a publicly available higher education dataset and demonstrates that Random Forest is the most effective algorithm for supporting accurate and reliable student academic status prediction systems.