TEISP: A TabNet-Based Framework Integrating Emotional Intelligence into Student Performance Prediction

Abstract Student performance prediction remains a significant challenge in Educational Data Mining, particularly when emotional and behavioral factors are considered alongside academic indicators. This study proposes a TabNet-based Emotional Intelligence Student Performance (TEISP) framework that integrates traditional academic attributes, including CGPA, attendance, and midterm scores, with emotional intelligence (EI) factors such as motivation, self-awareness, social skills, and stress management. The proposed framework employs TabNet, an interpretable deep learning architecture specifically designed for tabular data, capable of capturing complex nonlinear relationships while maintaining model transparency.Experiments were conducted using the VSIT–MIT dataset comprising 550 undergraduate students, 21 features, and three performance categories (Excellent, Average, and Poor). Model evaluation was performed using stratified 5-fold cross-validation to ensure robust and unbiased performance assessment. The proposed TEISP model achieved an average accuracy of 0.435 ± 0.018, with precision, recall, and F1-score values of 0.43, 0.42, and 0.42, respectively, and an AUC score of 0.88. Although the predictive accuracy is moderate, the model demonstrates superior stability, interpretability, and generalization compared with several baseline machine learning algorithms, including Naïve Bayes, K-Nearest Neighbors, Decision Tree, Random Forest, Logistic Regression, Support Vector Machine, and Gradient Boosting.Furthermore, SHAP-based explainability analysis reveals that emotional intelligence attributes, particularly motivation and self-awareness, play a substantial role in predicting academic outcomes. The findings highlight the importance of incorporating emotional and behavioral factors into student performance prediction models and demonstrate the potential of interpretable deep learning frameworks for educational analytics and decision support systems.