Tri-Attentive BiLSTM Framework for Integrated Student Performance Evaluation, Stress Detection, and Score Prediction

Abstract

Today’s students face a heavy academic workload, long study hours, and continuous pressure to perform well, which affects both their mental and physical health. Increasing stress, fear of failure, and lack of support often push vulnerable students toward severe outcomes, including depression and suicide. Using the provided student dataset containing features such as study hours, sleep duration, attendance, stress level, past scores, and extracurricular activity, we perform preprocessing and Exploratory Data Analysis (EDA) to understand behavior patterns and data quality. The existing system applies classical machine learning models like Random Forest, Support Vector Machine (SVM), Gradient Boosting, Decision Tree, and Linear Regression, but these handle each output separately and fail to learn deeper relationships between student factors. To address this, we propose an integrated Attention-based Bidirectional Long-Short Term Memory (BiLSTM) model combined with Optimal Decision Rule List (ODRL) feature extraction for joint prediction of student performance grade (classification), stress level (classification), and exam score (regression). The Attention-BiLSTM captures important behavioral sequences, while ODRL extracts meaningful features. Results show that the proposed approach outperforms the existing models, providing a more accurate and holistic understanding of student academic and emotional conditions.