This paper discusses the construction of thermal comfort prediction models with an unused machine learning model known as Fast Function Extraction. For problems such as Thermal comfort, it can address issues in Machine Learning such as the necessity for validation, hyperparameter tuning, noise sensitivity, and avoiding overfitting. It can reach accuracy close to other Machine Learning tools while maintaining the simplicity and performance characteristics of an algebraic formula.  The solutions constructed by this model are both equally as accurate and vastly quicker to calculate while being noise resistant, an area rarely discussed in the field of thermal comfort. Furthermore, there was little tuning or re-coding involved to produce such a result, and only one run for the dataset was needed as it is deterministic.

        In this paper, the research that has progressed over thermal comfort, as well as the methods and tools used to improve prediction, as well as the need for a reliable HVAC implementation, are discussed. The developed model has been tested against SVM in terms of accuracy as well as performance based on the RP-884 database. For training, it was close to SVM in lower data samples and outpaces it with higher data, showing high scalability. It was also found to outpace Support Vector Machine in prediction up to 900-fold, predicting thousands of samples in a millisecond. The conclusion is that this model is both practical and suitable to be implemented for Thermal Comfort prediction as a lightweight and efficient model for IoT.