Empirical Analysis of Machine Learning Methods for Predicting Student Performance

Performance prediction for students is relatively a new topic of research that aims to apply quantitative and qualitative approaches to evaluate student performance. This paper reviews existing approaches in predicting student performance across various learning environments. It explores fundamental statistical models, Machine Learning (ML) algorithms, and Deep Learning (DL) models, highlighting their pertinent characteristics and suitability for analyzing academic, behavioral and demographic data. The paper focuses on analyzing the existing methodologies for student performance prediction. The datasets used consist of student grades, attendance records, socioeconomic status, and interaction rates, along with an assessment of their efficiency in feature selection and data preprocessing. Lastly, the explainability and interpretability of the proposed predictive models are explored to enable reliability for educational stakeholders on these predictions for timely interventions. This survey concludes with a discussion of future scopes, focusing on enhancing prediction accuracy, ensuring fairness in models, and integrating multiple data modalities into performance analysis.