Analyzing the Determinants of U.S. Residential Energy Usage and Spending: A Machine Learning Approach
利用2020年居民能源消费调查数据和机器学习方法，识别影响美国居民不同能源类型及用途使用与支出的关键因素，发现CatBoost算法表现最佳，且各分析层次的重要特征不同。
This study explores the factors that impact residential energy usage and spending in the United States. Using data from the 2020 Residential Energy Consumption Survey (RECS), we investigate the significance of different energy consumption determinants at various analysis levels. Our analysis covers residential energy usage, electricity, natural gas, propane, and fuel oil consumption. We also examine energy usage for space heating, cooling, and water heating. To leverage the extensive RECS data, which includes over 180 variables, we utilized machine learning (ML) techniques for feature selection and determined their Shapley contribution for different target outcomes. Our results indicate that the CatBoost algorithm outperforms other ML techniques on the 2020 Residential Energy Consumption Survey sample. Our findings demonstrate that it is not appropriate to aggregate consumption and expenditure, as each level has distinct important features.