MACHINE LEARNING-BASED FRAMEWORK FOR PREDICTIVE ANALYSIS OF HUMAN THERMAL COMFORT IN INDOOR ENVIRONMENT

Abstract

The study aims to examine the potential of a machine learning framework in predicting the level of thermal comfort occupants is experiencing in an indoor environment. A Deep Neural Network (DNN) model was proposed and adopted to predict Fanger’s Predicted Mean Vote (PMV) and Predicted Percentage Dissatisfied (PPD) that indicates thermal comfort level. Six input variables affecting thermal comfort levels were obtained for this research to train and run the model.

The prediction results of PMV-PPD indexes obtained from the trained DNN model will be analysed and compared with the theoretical values derived using CBE Thermal Comfort Tool. Performance of the prediction model will be evaluated using 4 statistical metrics, namely, Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), Root Mean Square Error (RMSE) and Coefficient of Determination (R2). Although some large variations of errors were highlighted at specific data points giving the predicted PMV and PPD results with an MAPE of 29.48% and 19.60%, the proposed model is still able to obtain a maximum MAE of less than 5% and a high R2 of 0.89 and 0.78 respectively. With that being said, the proposed model generally shows a good ability to train and learn the data inputs efficiently and can obtain reasonable prediction results. Improvements can be made to the model for the prediction of PPD index. Recommendations given based on the results could serve as a guideline for future thermal comfort research. Limitation of the study is highlighted and could be considered in future studies.