2026-09-10T09:44:29Z https://drops.dagstuhl.de/oai

oai:drops-oai.dagstuhl.de:27574 2026-09-10T05:38:43Z ddc:004 open_access

From Feature Attribution to Spatial Information Structure: Interpreting Urban Pedestrian Flows with Explainable AI (Short Paper) Kiros, Atakilti Cohen, Achituv Ribakov, Yuri Klein, Israel GeoAI explainable artificial intelligence SHAP pedestrian mobility spatial information theory urban analytics feature attribution spatial data science Machine-learning models can predict pedestrian flows with high accuracy, but they provide limited insight into how urban environments organize movement. This paper introduces an interpretive framework that links explainable artificial intelligence with Spatial Information Theory by conceptualizing feature attribution as evidence of spatial information structure - the relative salience of temporal, morphological, accessibility, and environmental dimensions in shaping pedestrian movement. Using pedestrian sensor data from Zurich and Berlin, we train comparable CatBoost models and analyze SHAP-based feature contributions. We examine whether different urban contexts encode movement in systematically distinct ways. The results reveal clear contrasts: Berlin is dominated by temporal dynamics and situational conditions, whereas Zurich shows stronger influence from urban morphology and accessibility. We interpret these differences as model-based evidence that cities organize movement-related information differently and demonstrate that a compact SHAP-based comparative design provides a practical methodological bridge between GeoAI and the study of spatial representation. Schloss Dagstuhl – Leibniz-Zentrum für Informatik Atakilti Kiros and Achituv Cohen and Yuri Ribakov and Israel Klein 2026 Is Part Of LIPIcs, Volume 393, 17th International Conference on Spatial Information Theory (COSIT 2026) InProceedings Text doc-type:ResearchArticle publishedVersion application/pdf doi:10.4230/LIPIcs.COSIT.2026.30 urn:nbn:de:0030-drops-275744 https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.COSIT.2026.30 eng https://creativecommons.org/licenses/by/4.0/legalcode