Multiple Anomaly Alignments on Network Traffics

Anomaly subgraph detection has been widely used in various scenarios and fields (e.g., congestion related to passenger cars). Most existing methods for discovering anomalies in multiple attribute networks are embedding multiple attributes in a single network. The main challenge is that not all multiple networks can be integrated into a single network. It is challenging to optimize the same anomaly in multiple attributed networks. In this work, we propose the method for anomaly alignment across multiple attribute networks (A3MAN). Multiple anomalies are aligned with structural characteristics and anchor links between networks. Finally, we have conducted the experiments on the four real datasets, and the detected more aligned anomalies consisted of anchor links, which are 8.8 times that of the competitive methods (up to 63 times at the noise ratio 30%). For the case study on the real traffic dataset in Tianjin, China, we found that the “Yingfengdao” subway station located in the center of the “Nankai University Town” is the peak order location for bicycle-sharing and car-hailing region.