Explainable Machine Learning for Regional Load-Shedding Severity Classification Using Supply-Side Constraints in Bangladesh

An explainable three-class severity classification framework is proposed for nine divisions based on generation, supply-side constraints, weather data, calendar data, and regional characteristics. Various classifiers and a probabilistic soft Voting Ensemble approach are considered for severity classification in case of regional class imbalance. SHAP explanation provides insight into significant features and distinguishes the role of national generation versus regional supply-side constraints. A regional vulnerability analysis combines model explanations with observed load-shedding events to provide division-specific operational and policy insights.