Electricity Demand Forecasting: A Hybrid Approach using SARIMA and Gradient Boosting

Abstract

Electricity demand forecasting plays a critical role in optimizing energy generation, distribution, and planning. In this study, we present a data-driven approach combining time series analysis and Gradient Boosting models for short-term and long-term electricity demand prediction. The datasets used are three publicly available datasets from Kaggle, covering both production and demand metrics between 1985 and 2025. The proposed approach leverages feature engineering, statistical modeling (SARIMA), and machine learning (Gradient Boosting Regressor) to capture both seasonality and non-linear relationships. The final model achieved an RMSE of approximately 3956.11, demonstrating strong predictive capability. This research supports grid operators and policymakers by providing accurate, data-driven forecasts essential for improving grid stability and optimizing energy distribution.