AI-driven Predictive Analytics for Early Identification of At-risk Students Academically in Saudi Universities

Saudi Universities have rapidly adopted the digital learning platform as a part of a national approach for educational modernization, generating large volumes of academic, behavioural and interactional data.However, institutional monitoring mechanism is still relying largely on retrospective indicators, resulting in delayed academic interventions and underutilization of available data.Existing prediction models are often based on static machine learning methods that do not reflect the dynamic process of learning over time, do not account for the linguistic and cultural contexts inherent to Saudi education, and provide limited interpretability, reducing trust among faculty and administrators.To address these limitations, this research presents an AI-based predictive analytics framework utilizing a Contextual Temporal Fusion Hybrid Deep Network (CTF-HDN) with BERT4Edu contextual embedding to accurately represent features for semantic and behavioural feature representation and be optimized using the Crayfish Optimization Algorithm (COA).The model processes multi-source student data to analyse sequential learning behaviour and contextual educational factors, while offering interpretable insights through SHAP-based explanations.The Experimental evaluation on a real university dataset obtains 95.6% accuracy and 96.6% F1-score, supporting early risk identification and enhanced student retention aligned with Saudi Vision 2030.