An anomaly detection approach from spatio distributions of epidemic based on adjacency constraints in flow space

Shi, Y.; Wang, D.; Chen, Y.; Chen, B.; Zhao, B.; Deng, M.

In view of the limitations of existing methods for detecting potential epidemic spatial anomalies caused by multiple driving factors, this paper proposes a spatial anomaly detection approach for epidemic distributions constrained by crowd flow similarities. Firstly, those epidemic attributes that are significantly associated with crowd outflow intensity from the spread center are identified using the geographic detector. Then, considering all pairs of spatial units, a spatial weight matrix is adaptively constructed by measuring the similarity of crowd outflow intensities from the spread center. Finally, each spatial unit is characterized using the local variation gradient of epidemic attribute values, based on which both global and local Moran's I are modified to statistically discriminate the distribution patterns and detect local anomalous regions in flow space. Through performing comparative experiments on the spatio-temporal sequence of COVID-19, it illustrates that the proposed method can effectively detect the spatial anomalies caused by a variety of multiple potential factors. These findings can support the targeted epidemic prevention and control in different stages. © 2021, Surveying and Mapping Press. All right reserved.