Forecasting Hourly Electrical Load Demand: A Comparative Study of XGBoost and Bi-LSTM Models

The growing consumption of electricity, driven by population expansion, economic growth, and technological advancements, has increased the need for accurate demand forecasting to support effective energy planning. This study presents one machine learning model, eXtreme Gradient Boosting (XGBoost), and one deep learning model, Bidirectional Long Short-Term Memory (Bi-LSTM), for predicting hourly electrical load demand using hourly electricity consumption data and meteorological parameters from Vermont, USA (Balancing Authority: ISNE, Subregion: 4003) spanning 2018 to 2025. Model performance was evaluated using Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), the coefficient of determination (R²), etc. The results show that the Bi-LSTM model outperformed the XGBoost model by achieving lower MAPE, MAE, and RMSE values. While the XGBoost model performed well for very short-term forecasting of up to 6 hours, its overall accuracy remained lower than that of the Bi-LSTM model. In contrast, the Bi-LSTM model demonstrated strong performance for both short-term and long-term predictions, maintaining high accuracy for forecasting horizons of up to 24 hours. These findings demonstrate the effectiveness of the proposed machine learning and deep learning approaches in forecasting electrical load demand with satisfactory accuracy, thereby supporting electricity generation planning and sustainable energy growth.