Abstract

Indoor microclimate varies based on an ambient environment, significantly impacting occupant thermal comfort and health, which is particularly crucial in healthcare settings. Measuring the microclimate is challenging due to dynamic environmental factors such as temperature, humidity, and airflow. Additionally, building properties affect multi-indoor microclimates and the cost and complexity of installing sensors. This study addresses this concern using a machine learning to predict indoor microclimate based on thermal comfort. Our proposal explores the optimal position of a one-point reference that can represent and predict variations of multi-indoor microclimates throughout space, focusing on an inpatient ward measuring 9m × 9m × 3.2m. The comparative analysis is arranged to assess the predictive abilities of regression-based machine learning algorithms with hyperparameter optimization, including Artificial Neural Networks (ANN), Support Vector Machine (SVM), Decision Trees (DT), and K-Nearest Neighbors (KNN). Results show DT outperforming, with R-squared (R²), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE) values of 99 %, 0.09, and 0.19, which is precise and reliable for a multi-indoor microclimate prediction based on one-point reference instead of multiple devices. Key findings show that placing sensors in the most isolated location of the room provides the most effective monitoring of dynamic microclimates, as these positions experience minimal thermal interference and best represent conditions affecting vulnerable patients. This approach offers a practical and cost-effective solution for predicting indoor microclimate, enabling building managers to proactively respond to changing environmental conditions and improve thermal comfort, especially for vulnerable occupants in healthcare facilities, care homes, and educational institutions.