LAUSR

The large-scale data collected by automated fare collection (AFC) systems provide opportunities for researching both individual travelling behavior and mobility pattern in urban city. In existent researching files, AFC data focuses on detecting riders’ travel patterns. However, we leverage such data for detecting suspects exhibiting abnormal behaviors, such as “suspected theft”, “suspected begging” and “suspected unauthorized advertisement”. This paper first explores irregular data of “In-Out” in the same subway station (IOSSS) to extract the anomaly behavior of subway riders. IOSSS is defined as the riders entering and leaving via the same subway station. Then, it proposes a novel spatiotemporal feature map for those abnormal behaviors based on user clustering and ride behavior analysis. Finally, it proposes a spatial attention module+dense connected convolutional network (SAM+DenseNet) framework to distinguish abnormal suspects from regular passengers. The experimental results show the effectiveness of our proposed approach, with a precision value of 93.1%, a recall value of 97.3% and an F1 value of 95.2%. The visualization of the attention module could help police and public safety departments understand the features of abnormal behavior. Findings from this research can assist police and public safety departments in the city in taking proactive actions to track down suspects exhibiting abnormal behaviors.