Data Fusion Application in Predicting Human Comfort

Abstract

This paper studies the use of wearable device data along with other parameters such ambient temperature to model human comfort. Several machine-learning methods were used to build thermal comfort models from five individual’s wearable biometric data and their surrounding ambient conditions. The effects of the machine learning and input feature type and the output class size on the model accuracy were investigated. It is the goal to determine exactly what combinations of these factors will be able to accurately predict human thermal comfort. The accuracy of these models was determined by comparing their prediction to the individual’s actual thermal comfort found using voting input