Machine Learning Maps Why Traditional Villages Cluster Across Southwest China

Researchers have combined two statistical approaches to explain why thousands of China’s officially recognized traditional villages are scattered unevenly across the mountainous southwest, according to a study published in Humanities and Social Sciences Communications. The team analyzed the locations of 2,177 traditional villages and found that they are not distributed at random but form significant hierarchical clusters, with the densest concentrations in northwestern Yunnan and southeastern Guizhou.

The study integrated Multi-scale Geographically Weighted Regression, known as MGWR, with a machine learning method called XGBoost paired with SHAP, or SHapley Additive exPlanations. MGWR is designed to detect spatial non-stationarity, meaning that the strength of a factor’s influence can change from one geographic scale or location to another. XGBoost-SHAP, by contrast, captures nonlinear relationships and ranks the relative importance of each factor. Running both models on the same dataset allowed the researchers to cross-check local spatial variation against global, data-driven importance rankings.

The analysis showed pronounced spatial non-stationarity in how the influencing factors operate across different geographical scales, indicating that a variable that strongly shapes village locations in one area may behave differently elsewhere. Among all factors examined, the density of intangible cultural heritage emerged as the primary influence, carrying the highest SHAP importance value in the machine learning model. Cultural, environmental, and accessibility factors also displayed complex nonlinear relationships and synergistic interactions, meaning their combined effects on village distribution cannot be captured by simple linear models.

The authors state that this integrated methodological framework advances heritage science by providing interpretable machine learning tools for spatial pattern analysis. Because the models reveal both where influences vary and which factors matter most, the findings are intended to offer a scientific foundation for differentiated conservation strategies in mountainous regions, rather than a one-size-fits-all approach to protecting traditional villages. The study is published as an open access article, and the version shared ahead of final publication is citable and carries a permanent DOI, subject to further editorial edits before the Version of Record appears.

Subject of Research: Spatial analysis of traditional villages in Southwest China using MGWR and XGBoost-SHAP

Article Title: Spatial distribution patterns and influencing factors of traditional villages in southwest China: an integrated MGWR-XGBoost approach

Article References: Sun, Y., Gao, C., & Yang, Q. (2026). Spatial distribution patterns and influencing factors of traditional villages in southwest China: an integrated MGWR-XGBoost approach. Humanities and Social Sciences Communications. https://doi.org/10.1057/s41599-026-09116-5

Image Credits: AI Generated

DOI: 10.1057/s41599-026-09116-5

Keywords: traditional villages, Southwest China, spatial distribution, MGWR, XGBoost, SHAP, heritage science, spatial non-stationarity, intangible cultural heritage, Yunnan, Guizhou, conservation