Explainable Artificial Intelligence (XAI) for Interpretable Heritage Building Maintenance Prediction

Abstract

Heritage buildings are invaluable cultural assets that require continuous, informed care to prevent irreversible deterioration. Traditional maintenance prioritization relies heavily on expert judgment and periodic inspections, which are often subjective, reactive, and lack integration with quantitative environmental data. This paper introduces a novel Decision-Support Framework (DSF) that bridges the gap between predictive analytics and practical conservation workflows. The framework integrates an eXtreme Gradient Boosting (XGBoost) machine learning model, which predicts maintenance priority ratings based on historical microclimate data, with an interactive dashboard built on Microsoft Power BI. A key innovation is its emphasis on explainability, using SHapley Additive exPlanations (SHAP) to interpret model outputs, and visual analytics to transform complex predictions into actionable insights. Using a case study of Bangunan Stesen Keretapi Johor Bahru, a heritage railway building in Malaysia, we demonstrate the framework’s efficacy. The XGBoost model achieved a high F1-score of 0.88, and expert validation confirmed that the integrated dashboard significantly improved interpretability, transparency, and the speed of maintenance decision-making. This study underscores the potential of explainable AI and visual analytics to enable a paradigm shift from reactive to proactive and evidence-based preventive conservation.