Gender-specific thermal sensation analysis and SHAP-based prediction across steady, ramp-change, and step-change thermal conditions

Steady-state, ramp-change, and step-change thermal conditions commonly occur indoors and outdoors. However, comprehensive comparisons of different conditions’ effects on thermal perception remain limited, especially regarding gender differences. This study examined thermal perception and skin temperature under three conditions. Gender differences were assessed using independent t-tests and Mann–Whitney U tests. Personalized and generalized thermal sensation models were developed using machine learning, followed by SHAP analysis. Results showed: (1) Females felt cooler than males in neutral and cold environments at a metabolic rate of 1.1 met, with significant differences across steady-state, ramp-change, and step-change thermal conditions (p = 0.026* to p < 0.001**, Cohen's d = 0.338–0.830), while no significant gender differences were found in hot environments. (2) Chest skin temperature (T_chest) increased under cold exposure (17 °C). T_chest showed significant gender differences within the first 5 min after the temperature step-down to 17 °C, while back of hand skin temperature (T_{back of hand}) showed no difference. (3) Personalized models achieved high accuracy (94.63 % males, 95.24 % females), outperforming generalized models (74.07 % males, 77.09 % females). (4) The personalized model relied more on physiological data, while the generalized model favored environmental parameters. In stable-occupant scenarios, the personalized model can be embedded in wearables for individual thermal control. For scenarios with frequent occupant changes, the generalized model fits better. (5) T_{back of hand} was the most critical skin temperature for males’ thermal sensation prediction. Unlike males, T_chest contributed significantly to females. These findings contribute to gender-specific thermal sensation analysis and prediction.