LAUSR

Tree-based ensemble learning has received significant interest as one of the most reliable and broadly applicable classes of machine learning techniques. However, thus far, it has rarely been used to model and evaluate the drivers of energy consumption in buildings and as such its performance and accuracy in this field have yet to be properly tested or fully understood. The goal of this paper is to evaluate the performance of three ensemble learning algorithms in modelling and predicting the heating and cooling loads of buildings, namely (i) random forests, (ii) extremely randomized trees (extra-trees), and (iii) gradient boosted regression trees. Results show that the tested algorithms outperform the ones proposed in the recent literature, with gradient boosting improving on the prediction accuracy of the second best-performing algorithm by an average of 14% and 65% for the heating and cooling loads, respectively.