Predictive analytics in education: machine learning approaches and performance metrics for student success – a systematic literature review

Higher education institutions rely on student performance to improve grades and enhance academic outcomes. Universities face challenges in evaluating student achievement, providing high-quality instruction, and analyzing performance in a dynamic and competitive context. However, due to limited research on prediction techniques and the critical factors influencing performance, making accurate forecasts is challenging. The utilization of educational data and machine learning has the potential to improve the learning environment. Ensemble models in educational data mining enhance accuracy and robustness by combining predictions from multiple models. Approaches such as bagging and boosting effectively mitigate the risk of overfitting. Machine learning techniques, including Support Vector Machines, Random Forests, K-Nearest Neighbors, Artificial neural networks, Decision Trees, and convolutional neural networks, have been employed in performance prediction. In this study, we examined 85 papers that focused on student performance prediction using machine learning, data mining, and deep learning techniques. The thorough analysis underscores the importance of various factors in forecasting academic performance, offering valuable insights for improving educational strategies and interventions in higher education contexts.