Predicting Residents Thermal Comfort in Rural Mud Houses Using Machine Learning Classifiers Al

Business Agriculture and Technology, Dhaka, Bangladesh tafhimul.onedesk@gmail.com Thermal comfort in rural, naturally ventilated mud houses plays a vital role in enhancing productivity, general well-being, and climate-resilient housing. This study examines the advanced prediction models used through Artificial Intelligence (AI) techniques to predict residents’ thermal comfort based on both in-situ measurements and surveys. Emphasizing interaction between occupants and their surroundings, focusing on natural ventilation instead of mechanical systems, and employing SHAP (SHapley Additive exPlanations) to identify the most significant predictors of perceived comfort. A dataset of 500 residents from Kapasia, Gazipur, Bangladesh, including 24 environmental, physical, and demographic features (e.g., indoor/outdoor temperatures, humidity, indoor CO₂, room area, room height, age, gender, etc.) was collected through digital devices and questionnaires. The models (i.e., Random Forest (RF), Decision Tree (DT), and XGBoost) were trained using GridSearchCV hyperparameter tuning and cross-validation. Accuracy, Precision, Recall, F1-score and classification report were also used for evaluation. Each of the three ML models achieved an accuracy of over 80%, demonstrating their strong classification performance. According to SHAP analysis, age as well as outdoor temperature, and outdoor wall temperature have the greatest impact on how comfortable people feel, highlighting the crucial role that envelope thermal behavior plays in these homes. These findings contribute to explainable thermal-comfort modeling for rural mud houses and provide useful guidance for passive design, material selection, and ventilation strategies to support resilient, occupant-centered housing in low-income tropical contexts. Comparative Evaluation of Housing Delivery Models in Gauteng Metropolitan Municipalities: