Using an ensemble machine learning methodology-Bagging to predict occupants' thermal comfort in buildings

In this paper, an intelligent ensemble machine learning (EML) method - Bagging was developed for thermal perception prediction. Field data was collected in naturally ventilated (NV) and split air-conditioning (SAC) dormitory buildings in hot summer and cold winter (HSCW) area of China during the summer of 2016. The indoor physical measurement and subjective survey were conducted simultaneously during the field study. To determine the merit of the proposed Bagging approach, the performances of Bagging approach were compared against the artificial neural network (ANN) and support vector machine (SVM) regarding conventional statistical indicators, i.e., mean absolute error (MAE), root mean square error (RMSE), coefficient of determination (R-2) and Pearson correlation coefficient (r). Several thermal indices, i.e., thermal sensation (TS), effective temperature (ET*), standard effective temperature (SET*) and predicted mean vote (PMV), were adopted to access the occupants' thermal comfort and evaluate the model prediction. In our case study, the Bagging model's R 2 for TS, PMV, ET degrees and SET* were 0.4986, 0.9892, 0.9920 and 0.9900, respectively. It shows higher accuracy than SVM and ANN models in thermal perception prediction and outperforms the classical PMV index in TS prediction. Results indicate the proposed Bagging model's prediction performance is reliable and is highly accurate to predict the thermal perception. (C) 2018 Elsevier B.V. All rights reserved.