A Comparative Study on Machine Learning Algorithms for Assessing Energy Efficiency of Buildings

Abstract

An increase in energy demand in buildings continues to give rise to air pollution with a consequent impact on human health. To curb this trend, energy efficiency assessment plays a crucial role in helping to understand the energy in buildings and to recommend strategies to improve efficiency. Unfortunately, many existing approaches to assessing the energy efficiency of buildings are failing to do it accurately. Hence, the recommended energy efficiency strategies thereafter are failing to achieve the expected result. One approach in recent times uses data-driven predictive analytics techniques like machine learning (ML) algorithms to assess a building's energy efficiency towards improving its performance. However, as many ML algorithms exist, the selection of the right one is important for a successful assessment. Unfortunately, many of the existing works in this regard have simply adopted an ML algorithm without a justified rationale which may result in poor selection of the good performing ML algorithm. Therefore, in this study, a premise to compare the performance of ML algorithms for the assessment of energy efficiency of buildings was proposed. First, consolidated energy efficiency ratings of buildings from different data sources are used to develop predictive models using several ML algorithms. Thereafter, identification of best performing model was done by comparing evaluation metrics like RMSE, R-Squared, and Adjusted R-Squared. From the comparison, Extra Trees predictive model came top with RMSE, R-Squared, and Adjusted R-Squared of 2.79, 93%, and 93% respectively. This approach helps in the initial selection of suitable and better-performing ML algorithms.