Optimizing Grid Load Forecasting with Machine Learning

Accurate short-term load forecasting is critical for grid stability and cost-efficient energy dispatch. This research compares classical statistical baselines (ARIMA, exponential smoothing) against modern ML approaches (XGBoost, LSTM networks) across two years of regional consumption data. Key findings: - Gradient-boosted trees outperformed ARIMA by 18% MAPE during peak hours. - Feature engineering (weather, calendar, lagged demand) mattered more than model choice. - A hybrid ensemble delivered the most stable results across seasons. The full methodology, dataset description and reproducible notebooks are linked below.