Explainable Machine Learning for Housing Price Determinants: Focusing on Regional and Temporal Variation

The hedonic price model has long been a key method for predicting and explaining housing prices. Despite its widespread application, the model is limited by assumptions such as the normality of error terms and challenges like multicollinearity, which constrain its flexibility and predictive accuracy. Recently, machine learning techniques have emerged as powerful alternatives, demonstrating significant predictive capabilities while overcoming the limitations of traditional parametric models. Moreover, the advent of explainable machine learning methods has broadened their applicability across various domains, including housing price analysis. However, most existing studies on housing prices using machine learning have mainly focused on improving predictive performance with a limited set of variables, primarily emphasizing physical housing characteristics while overlooking the complex impact of locational factors. This study addresses this gap by leveraging actual transaction prices of apartments in Seoul between 2016 and 2018 to develop explainable machine learning models, specifically Random Forest and XGBoost. These models incorporate an extensive set of variables, including neighborhood urban environmental factors to provide a more comprehensive understanding of housing price determinants. Furthermore, this study examines how the importance of these variables and their influence on housing prices vary across different regions and time periods in Seoul, suggesting valuable insights into the dynamic and local context nature of housing markets