Student academic failure remains a persistent challenge in higher education, particularly in developing countries where late identification of at-risk students limits timely intervention. Recent advances in Educational Data Mining and Learning Analytics have enabled predictive modelling of student performance; however, many existing models suffer from poor interpretability, data imbalance, and limited integration of behavioral and socio-economic variables. This study presents a comprehensive review and synthesis aimed at guiding the development of an enhanced algorithm for student failure rate analysis. A systematic review methodology was employed, involving structured literature collection, screening, categorization of predictive techniques, and comparative analysis of statistical, machine learning, ensemble, and deep learning approaches. Algorithms were evaluated using established performance metrics including accuracy, precision, recall, F1-score, and ROC-AUC, alongside qualitative criteria such as interpretability, scalability, and real-time applicability. The analysis reveals that while ensemble and deep learning models achieve superior predictive accuracy, they often lack transparency and struggle with imbalanced educational datasets. Based on these findings, the research proposes a hybrid and explainable predictive framework that integrates ensemble learning, neural networks, imbalance-handling techniques, and explainable AI methods. The review demonstrates that hybrid approaches provide the most promising balance between accuracy, interpretability, and early detection capability. The major contribution of this research lies in synthesizing fragmented literature into a unified framework for enhanced student failure prediction, identifying critical research gaps, and establishing a methodological foundation for developing a scalable, interpretable, and real-time predictive system to support data-driven academic interventions.
Usman et al. (Thu,) studied this question.