ML ‐ SAS : A Machine‐Learning‐Based Spatial Autocorrelation Statistic for Mixed‐Type Data
Geographical Analysis · 2026-10 · 58 (4)
DOI: 10.1111/gean.70062 ↗Kurz gefasst
Fachartikel in der Zeitschrift Geographical Analysis, online erschienen am 09.10.2026. Verfasst von Woohyung Kim und 1 weiteren Autor:innen. Die Arbeit nennt 33 Literaturangaben.
Abstract (Originalsprache)
Existing measures of multivariate spatial autocorrelation often rely on covariance structures or fixed distances that are difficult to apply to nonlinear and mixed‐type attributes. This paper proposes the Machine‐Learning‐Based Spatial Autocorrelation Statistic (ML‐SAS), which combines a spatial weights matrix with an attribute‐similarity matrix learned from unsupervised random‐forest path proximities. Global and local statistics are defined within a generalized matrix‐correlation framework, with analytical randomization moments as theoretical references and rank‐based permutation tests for inference. Simulations assess correspondence with classical statistics, sensitivity to forest specifications and proximity definitions, null calibration, and local detection under continuous and nonlinear mixed‐type processes. Path‐based ML‐SAS was stable across the tested forest settings and less sensitive than terminal‐node proximity. Under the spatial null, the empirical probability of any rejection was 0.040. In a continuous multivariate design, ML‐SAS was more selective than Multivariate Local Geary. In a nonlinear mixed‐type design, adding a nominal attribute substantially increased recall and F1 relative to continuous‐only ML‐SAS, and leave‐one‐variable‐out analysis identified that attribute as the largest contributor to changes in the local similarity surface. In Dutch municipalities, adding ordered urbanity changed 23 local classifications. ML‐SAS provides a complementary framework for jointly analyzing continuous and categorical spatial attributes.
Zitieren
APA
Kim, W., & Lee, S. (2026). ML ‐ SAS : A Machine‐Learning‐Based Spatial Autocorrelation Statistic for Mixed‐Type Data. Geographical Analysis, 58(4). https://doi.org/10.1111/gean.70062
Vancouver
Kim W, Lee S. ML ‐ SAS : A Machine‐Learning‐Based Spatial Autocorrelation Statistic for Mixed‐Type Data. Geographical Analysis. 2026;58(4). doi:10.1111/gean.70062
BibTeX
@article{Kim2026,
author = {Kim, Woohyung and Lee, Sang‐Il},
title = {ML ‐ SAS : A Machine‐Learning‐Based Spatial Autocorrelation Statistic for Mixed‐Type Data},
journal = {Geographical Analysis},
year = {2026},
volume = {58},
number = {4},
doi = {10.1111/gean.70062}
}
EUVIMED