Leveraging Explainable AI to Enhance Student Metacognition Through Early Risk Detection

Aim/Purpose: Early identification of at-risk students is a burning issue in higher education. Although conventional Machine Learning (ML) models have high predictive accuracy, they tend to be opaque black boxes and offer no diagnostic information. This paper aims to fill this diagnostic gap by developing an eXplainable AI (XAI)-based framework to convert technical risk scores into actionable prompts for students’ self-regulation and reflection. Background: The research introduced a novel human-focused framework that fills the gap between predictive analytics and pedagogical theory. It shows how XAI can help turn a technical risk score into a metacognitive prompt, encouraging data-driven conversations between educators and students and offering a clear roadmap for institutional interventions. Methodology: Three ML models, Logistic Regression, Random Forest, and XGBoost, were used to analyze a refined sample of 278 student records. The methodology entailed a preprocessing pipeline of data. SHAP (Shapley Additive Explanations) was incorporated to support both global and local interpretability, whereas a formal Fairness Audit was performed to guarantee that risks were fairly detected across gender groups. Contribution: The research introduced a novel human-focused framework that fills the gap between predictive analytics and pedagogical theory. It shows how XAI can help turn a technical risk score into a metacognitive prompt, encouraging data-driven conversations between educators and students and offering a clear roadmap for institutional interventions. Findings: The analysis revealed that the Random Forest model achieved 92.9% accuracy and an AUC-ROC of 0.977. SHAP analysis found school absenteeism and mid-term grades as the most important risk predictors. Individualized diagnostics (waterfall plots) in the system provided the necessary evidence through student self-reflection, and the audit of fairness ensured that the model supports gender groups equally. Recommendations for Practitioners: Educational institutions must implement risk prediction systems based on XAI to go beyond mere warning systems. Practitioners should employ individual-level diagnostics to tailor mentoring and motivate students to reflect on their learning behaviors through evidence-based self-reflection. Recommendation for Researchers: Further studies are expected to include longitudinal pilot studies that quantify the actual behavioral effects of XAI-based prompts on student outcomes. Researchers are also advised to test the framework using larger, multi-institutional datasets to increase its generalizability. Impact on Society: Transparent and fair ML systems can improve student retention and graduation rates, leading to better resource allocation and a more inclusive educational environment. By focusing on student agency, these systems foster a more successful, self-aware workforce that benefits society in the long term. Future Research: To establish the global applicability and ethical soundness of the XAI framework, future studies must explore real-time application of XAI to learning management systems and test the cross-cultural validity of behavioral predictors.