Energy Consumption Prediction of Residential Buildings Using Machine Learning: A Study on Energy Benchmarking Datasets of Selected Cities Across the United States

Energy consumption around the globe has been rising for many decades. A significant portion of this consumption occurs in residential buildings. Developing reliable methods to understand and predict energy use is essential in the global effort to become more sustainable. Many cities across the U.S. have mandatory energy benchmarking programs requiring large buildings to track and report their energy use. These openly available datasets have encouraged many researchers to study energy use and develop energy use prediction models. In this study, we employ Extreme Gradient Boosting, Random Forest, and Artificial Neural Network as three common Machine Learning methods to predict building energy use in eight U.S. metropolitan areas. By examining the models’ performance, we also evaluate and compare the datasets provided by the benchmarking programs and we investigate whether the openly available datasets provide adequate input variables for energy use prediction. Based on the results, suggestions are provided to enhance the datasets and further improve building energy use research.