Maintenance Priority Rating Prediction Based on Microclimate Conditions for Heritage Buildings

Heritage buildings require proactive maintenance strategies that account for environmental vulnerabilities and conservation priorities. However, traditional inspection-based methods are limited by subjectivity and delay. This study presents a machine learning-based framework that leverages microclimate data-temperature, humidity, precipitation, and wind speed-to predict maintenance priority ratings for heritage buildings. Using a case study of Bangunan Stesen Keretapi in Johor Bahru, Malaysia, historical climate data from the Copernicus Climate Data Store were paired with building condition assessments to train and evaluate five models. Among them, XGBoost achieved the highest predictive performance with an F1-score of $\mathbf{0. 8 8}$. Rule extraction from XGBoost enabled interpretable insights, enhancing trust and transparency for decision-makers. Although based on a single site, the methodology serves as a scalable template for future implementations in similarly exposed heritage contexts. The findings demonstrate the feasibility of integrating environmental intelligence into conservation planning, promoting data-driven, preventive maintenance of culturally significant assets.