SHAP and LIME-Based Interpretability of AI Models in Electrical Grid Loss Forecasting Using Machine Learning algorithms

With the increasing complexity of electrical distribution networks and the widespread integration of AI for predictive analytics, ensuring transparency in model decision-making has become crucial.This study proposes an explainable AI (XAI) framework that integrates SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) for interpreting machine learning models applied to electrical grid loss forecasting.Using real-time grid telemetry data, we train ensemble models (XGBoost, Random Forest) and deep neural networks to predict power losses across various distribution nodes.SHAP is used to provide global interpretability, revealing the most influential features contributing to losses, while LIME offers local interpretability for individual predictions.Our experiments, conducted on a regional utility dataset, demonstrate high predictive accuracy, with SHAP and LIME yielding actionable insights for engineers and grid operators.These insights enable root-cause analysis, anomaly detection, and informed decisionmaking.The framework fosters trust in AI predictions and supports smarter energy management strategies aimed at reducing technical losses.This work serves as a foundation for integrating XAI in power systems, paving the way for transparent and accountable AI applications in smart grids.