This study introduces a novel methodology for subjective thermal comfort assessment by combining three indices thermal comfort, thermal sensation, and thermal preference in a finely graded scale index, facilitating the classification of occupants’ perceptions regarding the thermal environment. The aim is to identify the permutation form of those dimensions to derive an indicator that serves as a comprehensive label for training predictive models based on physiological measurements and machine learning. The research methodology involves the validation of the proposed index with the application on data collected from 76 subjects across three distinct experimental campaigns in a controlled environment. Subjects were exposed to diverse thermal conditions ranging from 20 ℃ to 28 ℃, resulting in a total of 435 tests. Participants, equipped with wearable devices for physiological signals recording (electroencephalogram, heart rate variability, galvanic skin response, and skin temperature) experienced a controlled setting for studying human comfort. Standard comfort evaluation questionnaires were employed to gather thermal perception data. The proposed index was then applied to classify physiological data collected from subjects, demonstrating the potential for using this permuted approach to streamline the creation of predictive models based on multidomain metrics. This innovative methodology significantly enhances the accuracy of thermal assessments, by integrating thermal comfort, sensation, and preference into one label for machine learning models, the predictive accuracy is heightened, allowing for a more nuanced understanding of individuals’ thermal experiences. The results suggest a promising avenue for advancing personalized comfort models, facilitating the optimization of indoor spaces for enhancing well-being and satisfaction.