A1338
Title: Tree-based methods for spatially dependent data: A case study of housing prices in Taipei
Authors: Yi-Hung Kung - Fu Jen Catholic University (Taiwan) [presenting]
Abstract: With the rapid intensification of globalization and urbanization, housing prices have become a key indicator of economic and social development. Taipei, as the capital of Taiwan, exhibits complex housing price dynamics driven by multiple interacting factors, including transportation accessibility, environmental characteristics, building attributes, socioeconomic conditions, and the distribution of public facilities, resulting in substantial spatial heterogeneity. A spatial regression tree (SRT) framework is proposed to identify the key determinants of housing prices by integrating spatial dependence into nonlinear modeling structures. While conventional regression and tree-based methods effectively capture variable relationships, they often fail to adequately account for spatial autocorrelation and localized variation. To address this limitation, spatial information is incorporated through a spatial weight matrix, and a spatial lag mechanism is embedded within the regression tree framework, resulting in a geographically adaptive tree-based model. Empirical results show that the proposed SRT model effectively captures regional heterogeneity and outperforms traditional approaches in terms of predictive accuracy. The findings provide insights into the spatial mechanisms underlying urban housing markets and offer a flexible analytical framework for spatially heterogeneous data, with important implications for urban planning and housing policy.
