Attention-Guided Trivariate Sequential Intelligence for Academic Performance Inference and Student Stress Profiling

Abstract

The increasing adoption of data-driven technologies in education has created a growing demand for intelligent systems capable of accurately analyzing student performance and behavioral patterns. Educational datasets contain valuable information, including study habits, attendance, engagement, and assessment records, which can be utilized to predict academic outcomes and identify factors affecting student success. Conventional Machine Learning (ML) models, including Random Forest (RF), Support Vector Machine (SVM), Gradient Boosting (GB), Decision Tree (DT), Logistic Regression (LR), and Linear Regression (LinR), provide baseline predictive performance but often struggle to model complex nonlinear relationships, exhibit limited generalization, and lack interpretability. To address these challenges, this study proposes BARM, a hybrid framework that integrates Bidirectional Long Short-Term Memory (BiLSTM) with an Attention mechanism and rule-based models, namely the Optimal Decision Rule List Classifier (ODRLC) and Optimal Decision Rule List Regressor (ODRLR). The proposed framework simultaneously performs two classification tasks, FinalGrade and StressLevel, along with ExamScore regression through a dynamic model selection strategy that identifies the best-performing model based on evaluation metrics. A Flask-based web application enables real-time prediction, batch processing, and Exploratory Data Analysis (EDA) visualization. Experimental results demonstrate that BARM improves prediction accuracy, reliability, scalability, and interpretability, making it well suited for practical educational analytics and academic decision support.