Improving Hospital Building Thermal Comfort Prediction: A Machine Learning Approach

Thermal comfort has a significant impact on productivity and occupant well-being. It has a meaningful effect on patient recovery, satisfaction, and overall health in hospitals. Naturally ventilated hospital buildings (NVHBs) are particularly sensitive environments due to the absence of mechanical cooling systems, even though they offer vital Medicare services to people from all socioeconomic backgrounds. This study developed machine learning (ML) techniques to accurately predict thermal comfort in NVHBs. Field data were collected from 200 inpatients in a rural NVHB in Faridpur, Bangladesh, using surveys, environmental monitoring, and patient feedback. The key features were selected by recursive feature elimination with cross-validation (RFECV) to improve model performance, and class imbalance was corrected using a hybrid of the edited nearest neighbor (ENN) approach and the synthetic minority over-sampling technique (SMOTE) method. This study presents the performance of three ML models: random forest (RF), decision tree (DT), and extreme gradient boosting (XGBoost), trained and evaluated using accuracy, precision, recall, and F1-score. Here, accuracy is a widely used and intuitive metric in classification models. RF outperforms all other models with a 98% accuracy on this dataset. The accuracies of XGBoost and DT were 95% and 91%, respectively. RFECV identified optimal features, and the combination of SMOTE-ENN balanced misclassification rates across different thermal comfort predictors. These results highlight both the limitation imposed by imbalanced datasets and the strengths of ML-based, interpretable models that can reliably predict thermal comfort in NVHBs, offering practical guidance for climate-resilient hospital design and improved patient well-being in resource-limited settings.