A Comparative Study of Machine Learning and Deep Learning Algorithms for Student Dropout Prediction in Higher Education

Abstract

Abstract Student dropout in higher education represents a significant academic and economic challenge. This study investigates the effectiveness of machine learning and deep learning techniques for early identification of at-risk students. A two-stage experimental framework is proposed. In the first stage, three machine learning algorithms (Random Forest, Support Vector Machine, and XGBoost) are compared with two deep learning models (Deep Neural Networks and TabNet) using the original dataset. In the second stage, the impact of data balancing techniques, namely SMOTE and Borderline-SMOTE, is evaluated. Model performance is assessed using accuracy, precision, recall, and specificity. The results demonstrate that XGBoost consistently achieves superior performance across both imbalanced and balanced datasets, while data balancing techniques significantly improve recall, enhancing the detection of at-risk students. These findings provide valuable insights into the role of data balancing in improving predictive performance in student dropout prediction.