Analyzing energy performance in American low income households: A DataDriven approach with machine learning insights

Abstract

The energy burden within low-income households presents a significant societal challenge that necessitates a thorough understanding for developing effective and sustainable policies. However, very few studies analyze and model energy-related factors among low-income individuals. Thus, this study critically investigates the relationship between various buildings' characteristics and four distinct energy-related outputs (i.e., energy burden, annual energy cost, annual energy cost per square footage, and annual energy cost per square footage per person) for both low-income and non-low-income households across the United States, using six machine learning algorithms. Leveraging the American Housing Survey dataset, this study identifies factors impacting energy burden and consumption and highlights variables contributing to the energy burden disparity among these two household groups, compromising census division, number of floors within the unit, household income, federal poverty level threshold, and number of rooms in the unit for non-low-income households, and the number of persons aged 18 and over, census division, household income, number of rooms in a unit, and unit size in square feet for low-income households. Additionally, it introduces 48 models to predict energy-related outputs and evaluates the accuracy of models. To evaluate the accuracy of these models, two key metrics are employed: R2 (coefficient of determination) and RMSE (root mean square error), which provide insights into the models’ predictive accuracy and error rates, respectively. The study reveals that Gradient Boosting Regression and XGBoost predict outcomes for low-income and non-low-income households with higher accuracy and with fewer inputs, making them cost-effective and efficient models. The findings enhance our understanding of energy equity issues and equip policymakers with actionable insights to alleviate energy burden disparities and promote energy equity across diverse socioeconomic landscapes.