关键词: 犯罪嫌疑人, 移动轨迹, 时空同现, 异常检验, 时空分布

Abstract:

Spatiotemporal co-occurrence patterns represent subsets of different object-types whose instances are frequently located together in both space and time. Using movement data to mine and analyze spatiotemporal co-occurrence patterns among diverse criminal suspects not only can help us better understand those unusual moving behaviors and relationships of them, but also provide decision-making supports for police departments in key suspects monitoring or arresting. Therefore, such pattern is one of the most important and useful way for the geography of crime researchers and police officers to extract and comprehend the implicit knowledge in large police databases which hold a large amount of crime data with spatiotemporal information. Additionally, to some extent, mining spatiotemporal co-occurrence patterns can also assist the police departments to save the limited police resources and improve their efficiency of handling criminal cases. However, current methods for mining spatiotemporal co-occurrence patterns can hardly be applied to the geography of crime studies directly because the way of determining spatial and temporal prevalence thresholds is presently difficult and lack of objectivity. Thus, in this paper, a novel candidate spatiotemporal co-occurrence pattern mining model was first built based on the spatiotemporal status co-occurrence pattern and the minimum spatiotemporal participation rate. Then, a framework for mining spatiotemporal co-occurrence patterns among criminal suspects under the point distribution was given through combining our proposed model and generalized ESD test. Finally, based on the proposed framework, a real case study in a province of China was conducted with an amount of real trajectory data of two criminal type (fraud and theft). The result shows that our proposed method is feasible in mining and analyzing the spatiotemporal co-occurrence patterns among criminal suspects. Specifically, 219 candidate spatiotemporal co-occurrence patterns were discovered under the condition that spatial neighbor distance equals to 688 meters and temporal neighbor distance equals to 504 seconds, and 6 of them were identified as the spatiotemporal co-occurrence patterns under the condition that significance level equals to 0.05. Importantly, the spatiotemporal distributions of those detected spatiotemporal co-occurrence patterns are not only approximately consistent with the common sense that criminal activities are more common in non-agricultural production areas, but also conform to the basic viewpoints of routine activity theory. This research expands the application of spatiotemporal co-occurrence pattern mining method to the geography of crime studies, and the study result can play an important role for police departments in key suspects monitoring and police resources allocation and deployment.

Key words: criminal suspects, trajectory data, spatiotemporal co-occurrence, outlier test, spatiotemporal distributions