XAI for Financial Sustainability Assessment and Prediction in Higher Education Institutions

Financial sustainability is a growing concern for Higher Education Institutions (HEIs). Many institutions face structural financial pressures due to reduced public funding, volatile enrolment, and rising costs. Although institutions collect increasing amounts of financial data, sustainability assessments often rely on backward-looking accounting ratios and provide limited insight into future patterns. Rather than proposing a new machine-learning algorithm, this study integrates and empirically evaluates established, complementary analytical techniques within one framework: financial Key Performance Indicator (KPI) engineering, a Financial Sustainability Index (FSI), machine-learning prediction, relative sustainability classification, institutional clustering, anomaly detection, and Explainable Artificial Intelligence (XAI). Using the public Higher Education Statistics Agency finance collection, we constructed a longitudinal dataset containing 2207 provider-year observations and examined income diversification, funding dependency, research activity, growth, and capital investment. Supervised prediction was evaluated on 1880 longitudinal records using leakage-safe expanding-window temporal validation. Across 1387 out-of-time observations, the persistence baseline achieved the strongest performance, with an R2 of 0.868, an MAE of 1.332, and an RMSE of 2.462. The Random Forest model remained strongly predictive, achieving an R2 of 0.852, an MAE of 1.587, and an RMSE of 2.612, while demonstrating stable performance across 30 random seeds. Temporal relative-sustainability classification achieved a mean accuracy of 0.842 and a macro-F1 score of 0.835. Unsupervised analysis identified three institutional profiles, while Isolation Forest flagged 111 financially atypical provider-year observations. Once current-year FSI was excluded, XAI analysis identified tuition dependency, income diversification, unrestricted reserves, and premises costs as the indicators on which predictions relied most. Because the FSI is internally constructed and the classification categories are sample-relative, the framework is presented as a proof-of-concept approach for sustainability screening and benchmarking rather than a validated predictor of financial distress.