A Points of Interest Outlier‐Driven Spatial Extent Framework for Urban Building Function Identification

ABSTRACT

Urban building function identification is often challenged by spatial inconsistencies between points of interest (POI) distributions and building footprints. To address this issue, we propose a framework that integrates scale‐related POI outlier screening with selective adaptive spatial extent delineation. POI outliers are detected via local aggregation scale analysis, and adaptive extents are constructed for the selected outlier categories. Building functions are then identified using XGBoost based on spatial extents, POIs, and built‐environment features. Across seven categories, our method achieves F 1‐scores ranging from 0.715 to 0.958, outperforming fixed‐buffer (100–500 m) and density‐based baselines. An ablation study reveals that removing either component reduces the F 1‐score by 0.02–0.22. Feature importance analysis indicates that 16 adaptive extent features occupy 12 of the top 20 ranks. Furthermore, city‐scale validation demonstrates that 80% of regions achieved an accuracy exceeding 0.90. This framework demonstrates that linking scale‐related POI semantics with selective adaptive spatial structures effectively captures urban functional patterns.