Assessment of ML Techniques for Student Dropout Prediction: A Comparative Study

Student dropout is a significant issue in higher education, affecting institutional efficiency and individual student outcomes. Existing research on dropout prediction methods lacks comprehensive evaluation, robust cross-validation steps and fails to evaluate various ML models on diverse metrics, necessitating a thorough comparative analysis. To this end this study aims to assess various ML techniques for predicting student dropout, comparing their effectiveness and identifying key predictors. The work analysed four widely-used ML models LR, RF, DT, and SGT. Our evaluation metrics included accuracy, precision, recall, and the area under the ROC curve to ensure a comprehensive performance. Furthermore, accuracy of the models is assessed with application of cross-validation and without application of crossvalidation employed to ensure robust and reliable performance comparisons. The results indicate that LR and RF are the most effective models, achieving accuracies of 77.03% and 76.8%, respectively with highest precision and a robust recall.