“…Blackbox models that predict heating/cooling loads have been built by using different statistical methods: linear regression [14,18,19,21], artificial neural networks [10,12,14,[16][17][18][19][20], support vector machines [9,11,15,16], autoregressive integrated moving average [13], gradient boosting regression [14], random forest regression [14], k-nearest neigbours regression [14], kernel ridge regression [14], Bayesian ridge regression [14], and singular value decomposition [22]. They input weather variables such as outdoor temperature [9-12, 14-16, 18-21], humidity [6, 10-12, 14-16, 18, 19], solar irradiance [11,15,16,[18][19][20], wind speed [14,19,20], precipitation [14], sky clearness [19], and categorical indoor variables such as the state of occupancy [13,14,17,19]. In some cases, synthetically generated data from BPS tools were used in lieu of measured data from real buildings [15,16,18,19].…”