With the rising complexity of student data and the growing need for timely academic interventions, higher education institutions increasingly require intelligent systems to support student success. This research proposes a machine learning-based early warning and recommendation system designed to identify undergraduate students at academic risk and provide personalized strategies to improve their outcomes. A comprehensive dataset containing institutional, academic, and demographic variables was used to train and evaluate fifteen machine learning classifiers, where data from semester N were used to predict academic warning status in semester N + 1, Among them, ensemble and hybrid models such as Boosting with Grid Search, Stacking, and Random Forest outperformed traditional methods in terms of accuracy, recall, and area under the curve (AUC). The best-performing model, Boosting with Grid Search, achieved an accuracy of 89.7, a recall of 0.92, and an AUC of 0.92, demonstrating the strong predictive capability of the proposed approach. Feature selection techniques, including correlation-based and information gain approaches, revealed that academic performance indicators like program average (SCGPA), semester average (grade point average; GPA) and cumulative average (CGPA) were the most predictive features. Beyond early detection, the system incorporates a recommendation module that identifies root causes of academic risk—such as course mismatch and disengagement—and suggests tailored solutions. SHapley Additive exPlanations (SHAP)-based explainability was employed to enhance model transparency, allowing stakeholders to understand and trust the decision-making process. The proposed system offers dual benefits: it improves the efficiency of academic advising while enhancing student retention through data-informed interventions. This work demonstrates the value of integrating predictive analytics with explainable AI (XAI) to support student-centered educational practices.
Taqatqeh et al. (Fri,) studied this question.