Capturing the distance decay effect of amenities on housing price using explainable artificial intelligence

Abstract

Automated valuation models (AVMs) have been significantly enhanced by machine learning techniques, continuously improving accuracy and flexibility in data structure and statistical assumptions. However, the complexity of machine learning models also creates challenges in interpretability, often rendering them as 'black box' models, which complicates understanding and acceptability in practice. In this context, this study uses Shapley Additive exPlanations (SHAP), a prominent explainable artificial intelligence (XAI) technique, to explore the influence of housing characteristics on price valuation in Greater Sydney, specifically aiming to capture the non-linear distance decay effect of amenities on housing price, which has been underexplored in previous studies. The results of applying SHAP in this study reveal that location characteristics, represented by accessibility to amenities, are the most important features in housing valuation models. Furthermore, the study shows that the non-linear contribution to housing price is dynamic by distance and diverse by type of amenity, such as education, transportation, and community facilities.