The continuous adoption of digital learning platforms has resulted in the generation of large-scale and heterogeneous educational data. Conventional analytical approaches are often insufficient to process such complex datasets effectively. Accurate prediction of student performance is essential for improving learning outcomes and enabling timely academic interventions. This study proposes an AI-driven predictive analytics framework for predictive analysis in smart education systems.The proposed architecture integrates scalable data preprocessing, adaptive feature engineering, and a hybrid ensemble learning model to enhance prediction reliability. Unlike traditional models that rely mainly on examination scores, the framework incorporates attendance behaviour, engagement patterns, assignment trends, and historical academic performance. A dynamic feature selection mechanism is introduced to identify influential predictors and reduce model complexity.When compared with baseline machine learning approaches, our experimental evaluation shows that the presented method improves on accuracy, precision, recall, and F1-score. Evidence from our evaluations supports the view that AI and Predictive Analytics provide significantly improved capabilities for early identification of academically at-risk students; thereby, enabling data-driven decision making in education.