Individual thermal comfort prediction using ensemble transfer learning and physiological signals under different physical characteristics

Thermal comfort models often suffer from low accuracy at the individual level due to their reliance on population-based statistics and environmental factors. While physiological signals offer a personalized alternative, data collection across diverse demographics remains a significant challenge. This study proposes an ensemble transfer learning (TL) framework to predict individual thermal comfort by transferring knowledge between different gender groups. A hybrid model integrating 1D-convolutional neural networks (1D-CNN) for feature extraction and support vector machines (SVM) for classification was developed. The model was pre-trained on data from male subjects in their 30s (source domain) and fine-tuned for female subjects in their 20s (target domain) using a bagging-based ensemble strategy. The experimental results showed that the ensemble TL model outperformed other approaches, achieving a thermal comfort prediction accuracy of up to 97% for the female target group. This approach demonstrates that high-performance personalized thermal comfort models can be established even with limited data from specific demographic groups, providing a scalable solution for intelligent building energy management and occupant wellbeing.