Harnessing Explainable AI (XAI) to Inform Educational Decision-Making: A Machine Learning Study on Graduation Rates

Improving student outcomes and institutional performance is a central theme in Educational Data Mining (EDM), where machine learning (ML) models are increasingly used for data-informed policymaking. However, the lack of transparency and interpretability in many ML models limits their utility fornon-technical stakeholders. In this study, I used institutional-level data from the U.S. Integrated Postsecondary Education Data System (IPEDS) between 2012 and 2017 to predict graduation rates—a keymetric of institutional effectiveness. I applied six supervised ML algorithms, with Support Vector Regression (SVR) achieving the highest test R2 of 81.12%, followed closely by ensemble methods such asXGBoost, Random Forest, and LGBM. To enhance interpretability, I incorporated Explainable AI (XAI)tools including SHAP, Permutation Importance (PI), Partial Dependence Plots (PDPs), and IndividualConditional Expectation (ICE) plots. These analyses identified prior retention rates, equity gaps in graduation by race and gender, tuition levels, and test scores as key predictors. Force plots and waterfallcharts were used to provide local explanations of individual predictions. The study demonstrates howXAI can support transparent and actionable educational policy.