Land Price Dynamics: An Interpretable Machine Learning Approach

Land prices are a crucial aspect of urban planning and property valuation, and as cities grow, their dynamics become increasingly complex. Machine learning has been actively employed since the mid-2010s to capture these intricate relationships, but its black box nature poses interpretability challenges. To address these concerns, this study adopts an interpretable machine learning approach, employing the Extreme Gradient Boosting algorithm to estimate land prices in Gwangjin-gu, Seoul, and using Shapley Additive explanations values to interpret the results. This study uncovers nuanced relationships between various factors and land prices, including the nonlinear effects of the distance from amenities and disamenities, optimal road widths, and suitable land parcel sizes for both residential and commercial lots. By providing interpretations specific to the domain of urban planning, and highlighting factors unique to residential and commercial land markets, this study offers actionable insights that are expected to guide targeted urban planning strategies.