DESIGN AND VALIDATION OF AN OPTIMIZED HYBRID AI MODEL FOR EARLY ACADEMIC RISK PREDICTION AND LEARNING BEHAVIOR ANALYTICS

Abstract

 Rapid expansion of digital learning platforms has produced vast amounts of educational data to aid in identifying students at academic risk early. This study introduces a new hybrid Optimized Artificial Intelligence (AI) model that combines Convolutional Neural Networks (CNN), an Attention Mechanism, and Bidirectional Gated Recurrent Units (BiGRU) for the purposes of early academic risk prediction and learning behavior analytics. The model uses HEI, XAPI and HESP datasets and data pre-processing methods such as data cleaning, label encoding, Min–Max normalization, SMOTE and Recursive Feature Elimination (RFE). The CNN extracts meaningful features while the Attention Mechanism highlights the most relevant learning behavior indicators, and the BiGRU models the temporal dependencies to enhance prediction accuracy. The proposed framework is validated with the help of the metrics such as Accuracy, Precision, Recall, F1-score and AUC–ROC. The study offers an intelligent and reliable decision-support system that can help educational institutions to detect at-risk students in time and timely interventions to improve their academic success.