A Machine Learning Framework for Evaluating College Performance and Student Outcomes

The increasing reliance on machine learning (ML) for predicting college performance and student outcomes necessitates robust, scalable models that can generalize across diverse institutions. This study leverages the publicly available College Scorecard dataset to develop an advanced predictive framework using ensemble learning and deep learning techniques. Key preprocessing steps included handling missing values, detecting outliers, and feature engineering to enhance model interpretability. The proposed framework employs Random Forest, Gradient Boosting, and Neural Networks, with deep learning models achieving the highest accuracy of 89.3%. Explainable AI (XAI) methods, such as SHAP, were integrated to interpret model decisions, ensuring transparency. Compared to prior studies relying on limited datasets and traditional ML models, our approach demonstrates superior predictive performance and generalizability. The findings highlight the potential for data-driven decision-making in higher education, enabling institutions to implement personalized interventions and optimize student success strategies.