Enhancing Academic Performance Evaluation of Research Topic Selection through Explainable Artificial Intelligence and Educational Data Analytics

This research work presents a framework that leverages Explainable Artificial Intelligence (XAI) and Educational Data Analytics (EDA) to evaluate student performance in the various departments. A dataset comprising student-authored research articles, including titles, abstracts, citations, and online engagement metrics, was analysed using Logistic Regression (LR) for topic classification. The model achieved an overall accuracy of 93% with 7% miss rate, with detailed class metrics as follows: class 0 (Physics) having precision 1.00, recall 0.22, f1-score 0.36, support 37, class 1 (Engineering) having 0.90, 0.89, 0.89, 200, class 2 (Computational Science) having 0.93, 1.00, 0.96, 591, and class 3 (Pure Mathematics) having 0.00, 0.00, 0.00, 9. Macro-average precision, recall, and f1-score were 0.71, 0.52, and 0.55, while weighted averages were 0.92, 0.93, and 0.91. To make it more interpretable, SHAP and LIME analyses were performed, which revealed the textual features and tendencies of scholarly activity that produce the greatest impact on the results of classification. The results indicate that Physics students are always involved in submitting and publishing manuscripts, and this reflects on their academic performance throughout the academic calendar. Such an approach provides a great predictive power and practical information, providing the educator with a data-based tool that is transparent and allows tracking and improving student academic achievement.