As personal environmental control systems (PECS) gain prominence over conventional whole-space conditioning, accurately predicting personalized and localized thermal sensation (TS) becomes increasingly important. This study systematically evaluates the relative performance of two physiological predictors—local skin temperature and local heat flux (rate of heat dissipation from body)— in data-driven models of localized TS. Drawing on data from two human subject experiments that captured both physiological signals and subjective thermal responses, we address key research questions concerning (i) the predictive accuracy of skin temperature versus heat flux, (ii) regional variations in model performance across 16 body segments, (iii) the influence of different modelling strategies (i.e., separate models per subject/body part versus a unified model with individual-specific features), and (iv) model robustness when applied to an independent dataset. Results indicate that incorporating personalized and localized modelling strategies can reduce prediction error relative to non-personalized/localized approaches. Most body regions exhibited comparable performance between skin temperature- and heat flux-based models, while skin temperature-based models outperformed in extremities, such as the hand (error of 0.42 versus 0.56). We hypothesize that this performance difference can arise from the complex relationship between heat flux and TS in extremities due to the influence of vasomotor thermoregulation. Conversely, when testing model robustness on an independent dataset, heat flux-based models exhibit a lower error for the hand (0.36 versus 0.49). Findings establish heat flux as a viable alternative to skin temperature in predicting local TS. These advancements are crucial in optimizing the design and control of next-generation PECS.