Estimating Residential Natural Gas Demand and Consumption: A Hybrid Ensemble Machine Learning Approach

Abstract

The routine use of natural gas, particularly in residential settings, has been integral to human activities for many decades. This study proposes a hybrid ensemble regression machine learning model for forecasting residential natural gas demand. Accurate demand prediction is essential for efficient energy management and resource planning. The proposed approach integrates multiple regression algorithms including K-Nearest Neighbors (KNN), Support Vector Regression (SVR), Decision Tree Regression (DTR), and Linear Regression (LR) to leverage the strengths of each model and enhance overall predictive performance. The ensemble method operates in two phases: training individual regression models on the dataset, followed by aggregating their predictions. Model performance is evaluated using metrics such as Mean Absolute Error (MAE), Mean Squared Error (MSE), coefficient of determination (R²), and prediction accuracy, and is benchmarked against individual models. Cross-validation techniques were applied to ensure the robustness of the results. Experimental consequences demonstrate that the hybrid ensemble approach consistently outperforms standalone models by capturing diverse patterns and relationships within the data.