A Hybrid Classification and Clustering Approach for Multi-Class Student Performance Prediction and Profiling

Effective educational planning and immediate academic interventions in higher education are dependent upon accurate and clear analysis of the academic performance of a student. Additionally, this research offers a proper machine learning approach that prioritises systematic pre-processing, model selection, interpretability, and student profiling for individualised support. To guarantee data quality and strong model generalisation, extensive data pre-processing methods are first used, such as categorical encoding, missing value treatment, feature scaling, and class-stratified sampling. Then, using important academic and behavioural characteristics like attendance, study habits, and previous academic records, supervised classification models are used to predict multi-class academic performance. With regard to recall, accuracy, precision and F1-score, the Logistic Regression model outperformed with the 96% the ensemble-based classifiers, also achieving a good result of 92%. Permutation-based feature importance is used to find dominant factors influencing academic outcomes in order to improve interpretability. Additionally, unsupervised K-Means clustering is used to identify significant student groups, which are then verified by silhouette analysis and displayed using principal component analysis. The resulting profiles facilitate data-driven, targeted recommendations that are carried out via an interactive application built on Streamlit.