Student Dropout Risk Classification: A Comparison of Random Forest and Naïve Bayes Algorithms

Abstract

Student dropout challenges higher education, while class imbalance can obscure models’ ability to identify students at risk. This study compared Random Forest and Naïve Bayes using academic and nonacademic variables, emphasizing performance on the dropout class. Evaluation used 1,900 test records: 1,452 non-dropout and 448 dropout cases. Confusion matrices, classification metrics, a majority-class baseline, balanced accuracy, and Matthews correlation coefficient were examined. Naïve Bayes achieved 76.58% accuracy, exceeding Random Forest’s 73.21% but barely surpassing the 76.42% baseline. For dropout detection, Random Forest achieved 0.20 recall and a 0.26 F1-score, compared with 0.05 and 0.09 for Naïve Bayes. Random Forest identified 91 dropout cases versus 22, with approximately 1.93 additional false alarms per additional case detected. Despite its relative advantage in detecting dropout, Random Forest missed 357 of 448 cases. Neither model is suitable as a primary early warning classifier without further improved handling of class imbalance and careful decision thresholds.