Predicting Student Performance Using BiLSTM With Synthetic Institutional Data Mapping: A Case Study Contextual Integration of OULAD and AISHE in Educational Context

Student performance prediction has become an important research area in educational data mining because it enables early identification of academically at‐risk students and supports timely educational interventions. However, the development of robust prediction models for Indian higher education is constrained by the limited availability of comprehensive student‐level institutional datasets due to privacy and data‐sharing restrictions. To address this challenge, this study proposes a hybrid student performance prediction framework that integrates the Open University Learning Analytics Dataset (OULAD) with institutional characteristics derived from the All‐India Survey on Higher Education (AISHE) through a synthetic institutional mapping strategy. The proposed approach probabilistically associates OULAD learner records with representative Indian institutional profiles, enabling the incorporation of institutional context without requiring access to sensitive student data. A Bidirectional Long Short‐Term Memory (Bi‐LSTM) network optimized using the Adam optimizer is employed to model temporal learning behaviors across 32,593 student records. The feature engineering process integrates 21 predictive features, comprising 14 student‐level behavioral and academic attributes extracted from OULAD and 7 institutional‐level characteristics derived from AISHE. The proposed Bi‐LSTM architecture consists of 128 bidirectional LSTM units with a dropout rate of 0.3 to improve generalization. Experimental evaluation demonstrates that the proposed framework achieves an accuracy of 92.5%, precision of 90.2%, recall of 94.1%, and F1‐score of 92.1% for four‐class student performance prediction (Low, Medium, High, and Very High). Comparative experiments against Support Vector Machine, Random Forest, and conventional LSTM models demonstrate accuracy improvements of 8.3%, 6.7%, and 3.2%, respectively. Furthermore, SHAP analysis identifies assessment scores, assessment submission rates, Virtual Learning Environment engagement, and activity diversity as the most influential predictors, improving the transparency and interpretability of the proposed model. The proposed framework offers a scalable, reproducible approach for student performance prediction using integrated behavioral and synthetic institutional data.