This research proposes a predictive framework for assessing the academic performance of students in higher education, leveraging data mining and machine learning techniques. The study addresses the challenge of imbalanced datasets, which often skew the performance of classification models, by employing the Synthetic Minority Oversampling Technique (SMOTE) to enhance prediction accuracy. The framework integrates various supervised learning algorithms, including J48, Random Forest (RF), and K-Nearest Neighbors (KNN), to predict student performance and recommend suitable academic paths based on historical data. The data set, collected from the Higher Institute for Management Sciences, spans multiple academic years, and includes student records from three departments: Information Systems, Management, and Accounting. The research demonstrates that handling class imbalance through SMOTE significantly improves model performance, with the Random Forest classifier achieving the highest accuracy of 93.70%. The study also highlights the importance of feature selection, data preprocessing, and normalization in optimizing predictive outcomes. The findings underscore the potential of educational data mining to support early academic interventions and personalized guidance, aiding students in making informed decisions about their academic trajectories. Future work will explore additional sampling techniques and expand the dataset to further enhance model accuracy.