Toward explainable and scalable property valuation via BIM and GA-enhanced ensemble learning

Purpose This study proposes a novel BIM-integrated, explainable and scalable framework for property valuation. It aims to enhance predictive accuracy and transparency by combining IFC-based feature extraction with a genetic algorithm enhanced ensemble learning, addressing key limitations of traditional and opaque AI models in the built environment. Design/methodology/approach An IFC-based pipeline converts BIM geometry and semantics into machine-readable features. A genetic algorithm-enhanced gradient boosting regressor (GA-GBR) is trained and tested on 152 k transactions from China, and then further tested from building-type perspective at smaller scales. Hyperparameter optimization is performed using a genetic algorithm, and model interpretability is enabled through SHAP. Findings The GA-GBR model outperforms 11 benchmark models, achieving a 3–4.6% gain on MAPE over recent state-of-the-art methods. SHAP analysis identifies key predictors – local price level, transaction timing and floor area – while submarket results highlight context-specific drivers such as elevator presence in high-rise buildings. GA-based optimization enhances both predictive performance and feature relevance. Practical implications The framework supports automated, explainable and scalable valuation using BIM-derived features, enabling end-to-end deployment and informed decision-making. It offers valuers, designers and policymakers a transparent tool for assessing property value at multiple scales. Originality/value This is the first study to integrate GA-optimized ensemble learning, IFC-derived features and xAI techniques into a unified BIM-valuation workflow validated against real-world data. It contributes to methodological advancement while facilitating industry adoption of explainable AI applications in the built environment.