Global aging and intensifying heat extremes heighten outdoor thermal risk for older adults. Concurrent micrometeorological, physiological, and thermal-sensation data were collected from older park users in Xi’an, China, during sitting, slow walking, and square dancing. Predictors selected by statistical and machine learning methods were fed into gender-specific gradient boosting machine models of thermal sensation. Results indicated that: 1) Neutral physiological equivalent temperature (NPET) shifted markedly with activity. Older females showed lower NPET than older males during sitting, while during slow walking and square dancing, they exhibited higher NPET than males. 2) Older women maintained higher heart rates (HR) than men across all activities; while older men displayed a U-shaped core temperature (T_c) curve during slow walking. Limb skin temperatures (ST) exhibit more pronounced fluctuations with increasing activity intensity, whereas chest ST remained stable. 3) Thermal sensation vote (TSV) predictors differed by gender and activity. For men, ST dominated during sitting, HR during walking, and T_c during dancing; for women, TSV correlated strongly with ST_{upper arm} during sitting, while correlated jointly with T_c and ST_chest during square dancing. 4) The gender-specific TSV model, requiring only two to three localized physiological inputs, outperforms conventional indices in accuracy and feasibility. It directly informs age-friendly park design, enhances outdoor safety for older adults, and supports sustainable urban development.