Machine learning-based prediction and transformation of thermal sensation votes (TSV) under different scales for elderly people in summer

Predicting and transforming the Thermal Sensation Votes (TSV) under different scales plays a critical role in setting appropriate temperature setpoints of air conditioning for elderly people with different thermal preferences. This study aims to predict and transform the TSV under four different scales for elderly people. Firstly, the TSV data under four scales and environment parameter data are collected from two pensioners’ buildings. Secondly, five Machine Learning (ML) algorithms are evaluated by accuracy, recall, F1 score, AUC and ROC curve, without and with outliers being removed by the Edited Nearest Neighbour algorithm, to determine the optimal algorithm for the prediction. Lastly, the distribution ranges of environment parameters outputted by the optimal algorithm are used to conduct the transformation. Results indicate that the Adaptive Boosting shows optimal performance with all evaluation parameters exceeding 0.9, followed by the Random Forest, logistic regression, Artificial Neural Network and Naive Bayes. The mean air temperatures outputted by the AB algorithm are 31.2 °C and 34.0 °C for the two warm-side classes of the 5-point scale, 30.9 °C, 32.3 °C and 34.9 °C for the three warm-side classes of the 7-point scale, and 30.8 °C, 31.9 °C, 33.5 °C and 35.4 °C for the four warm-side classes of the 9-point scale. These temperatures indicate granular scales can capture subtle thermal sensation variations. This study provides method support for TSV scale transformation of elderly people and provides data reference for the indoor environment parameters of pensioners’ buildings. •The transformation of four TSV scales addresses the thermal needs of elderly people.•Five ML algorithms are evaluated by five parameters to enable the transformation.•The ENN algorithm is used to eliminate outliers when their presence is confirmed.•The performance of the five algorithms is as follows: AB > RF > LR > ANN > NB.•Granular TSV scale can capture subtle thermal sensation variation for elderly people.