Daily Residential Natural Gas Demand Forecasting Using Machine Learning Regression: Comparative Evaluation With a Case Study in Qazvin Province, Iran

ABSTRACT Accurate short‐term forecasting of residential natural gas consumption (NGC) is essential for operational planning and supply reliability. Most forecasting studies rely primarily on meteorological variables, often neglecting infrastructure expansion effects. This study introduces the subscription growth ratio (SGR) as a socioeconomic indicator to enhance daily residential NGC forecasting in Qazvin Province, Iran. Two datasets were developed: Dataset A containing meteorological variables and Dataset A + SGR incorporating subscription growth information. Forecasting performance was evaluated using expanding window cross‐validation across 52 sequential folds. Multiple regression models were implemented, including multiple linear regression (MLR), support vector regression (SVR), random forest regression (RFR), and XGBoost (XGB). Feature importance analysis confirmed that temperature variables dominate NGC variation; however, SGR ranked as the third most influential predictor, exceeding maximum temperature. Results show that incorporating SGR significantly improves nonlinear model performance, reducing MAE by approximately 20% in RFR and XGB and increasing adjusted R ² by about 38%. Paired hypothesis testing confirmed statistically significant improvements for SVR ( p < 0.001), RFR ( p = 0.000347), and XGB ( p = 0.012654), while MLR showed no significant improvement ( p = 0.767816). The findings demonstrate that infrastructure‐driven demand growth has a nonlinear influence on residential NGC and should be integrated into operational forecasting frameworks.