Fairness and Bias Mitigation in Student Success Prediction Models

The adoption of machine learning (ML) and artificial intelligence (AI) in higher education has revolutionized student success prediction by enabling institutions to forecast academic outcomes, identify at-risk learners, and provide timely interventions. Yet, these predictive systems often inherit historical and structural inequities present in educational data, leading to algorithmic bias and unfair treatment of marginalized student groups. This study proposes a fairness-aware predictive framework that integrates bias detection, mitigation, and interpretability into all stages of the modeling process. Using a real-world Student Performance & Behavior Dataset containing 5,000 student records, five ML algorithms, Random Forest, Gradient Boosting, Support Vector Machine, K-Nearest Neighbors, and Logistic Regression were evaluated based on accuracy, precision, recall, and F1-score, alongside fairness metrics to assess equity performance. The Random Forest model achieved the highest predictive accuracy (37%), but fairness analysis revealed class imbalance and unequal group representation. To address these disparities, the proposed framework emphasizes three guiding principles: Fairness by Design, incorporating fairness constraints during data preprocessing and model training; Ethical Transparency, employing explainable AI tools to ensure accountability and stakeholder understanding; and Sociotechnical Alignment, embedding algorithmic decisions within institutional equity policies. The results highlight the importance of harmonizing predictive performance with fairness and ethical accountability. Ultimately, this work advances the field of responsible learning analytics by demonstrating that bias-aware predictive modeling can transform student success systems from efficiency-driven mechanisms into equitable, transparent, and socially responsible tools that foster educational justice and inclusion.