Data-Driven Approach for Evaluating the Energy Efficiency in Multifamily Residential Buildings

Abstract

Cities account for more than 70% of global fossil fuel use and greenhouse gas emissions. This number is likely to increase due to urban population growth. Much of the energy used in cities is consumed in buildings (e.g., for space conditioning and lighting). Better understanding of energy use patterns therefore is paramount. This paper leveraged advances in machine learning to model energy consumption in residential buildings and gain insights into building energy consumption trends in Chicago. By merging demographic and socioeconomic data collected from the US Census Bureau with energy benchmarking data for Chicago, three models were developed using three different machine learning algorithms: back-propagation neural network (BPNN), extreme gradient boosting (XGBoost), and random forest (RF). The results showed that XGBoost better predicts the building energy use, with an accuracy of 68%. Furthermore, Shapley Additive Explanations (SHAP) was used to interpret the impact of each variable used on building energy consumption. Overall, the insights gained in this study can help policy makers and planners to address building energy use better.