Boosting the accuracy of property valuation with ensemble learning and explainable artificial intelligence: The case of Hong Kong

Abstract This paper proposes a novel three-level ensemble learning model to boost the accuracy of property valuation. Compared with current machine learning models, our ensemble learning strategies perform better. Explainable artificial intelligence methods are applied to identify significant housing price determinants. The results show that most features have nonlinear relationships with price. The property’s age and geographic coordinates are the most important features and the density and diversity of urban amenities positively relate to price. The number of building blocks or housing units shows a threshold effect on price, and a price premium of higher public transit accessibility is witnessed. Implications for future property development and urban planning are provided regarding building rehabilitation and renewal, property development preparation and urban center identification, land use planning and architectural design, integrated development of metropolitan amenities, and transit-oriented development.