Spatial Anomaly Identification utilizing an Adjacency-Weighted Spatial Outlier Factor (AWSOF)

Abstract

Spatial outlier detection remains a challenging problem because anomalous observations are 
inherently localized and strongly influenced by spatial coordinates and neighborhood 
structure. Conventional detection techniques are particularly susceptible to masking and 
swamping effects, whereby extreme neighboring observations either conceal true spatial 
anomalies or incorrectly classify normal observations as outliers. Furthermore, many existing 
approaches rely on directional neighborhood comparisons, such as left–right or unidirectional 
evaluations, limiting their ability to capture the multidirectional spatial relationships that 
characterize complex geographic patterns. To address these limitations, this study proposes 
a novel Adjacency-Weighted Spatial Outlier Factor (AWSOF) framework based on an 
Exponential Decay Weight (EDW) model that incorporates local topological adjacency into the 
construction of spatial proximity weights. The proposed weighting scheme quantifies the 
influence of spatial proximity on variations in behavioral attributes by assigning exponentially 
decreasing influence with increasing neighborhood distance. Absolute attribute differences 
are projected across coordinate-based neighborhood vectors to derive the AWSOF, thereby 
providing a robust multidirectional measure of spatial deviation while minimizing the influence 
of localized extremes. The performance of the proposed framework was rigorously evaluated 
through extensive simulation experiments and validated using multiple real-world spatial 
datasets. Comparative analyses against conventional spatial outlier detection methods 
demonstrate that the proposed exponential weighting framework consistently delivers 
superior detection accuracy, robustness, and reliability under varying spatial configurations. 
Empirical results reveal that the EDW-based AWSOF effectively mitigates masking and 
swamping effects while preserving neighborhood integrity, leading to more accurate 
identification of genuine spatial anomalies. In contrast, the standardized approach, which 
behaves similarly to an arithmetic mean, remains highly sensitive to extreme localized 
observations, resulting in distorted neighborhood influence and reduced diagnostic 
performance.