Fast Prediction for Suspect Candidates from Criminal Networks

Abstract

Machine learning approaches have been introduced to support criminal investigations in recent years. In criminal investigations, Criminal acts may be similar, and similar incidents may occur consecutively by the same offender or by the same criminal group. Among the various machine learning algorithms, network-based algorithms will be suitable to reflect such associations. In general, however, inference by network-based algorithms is slow when the size of data is large, so it is fatal in crime scenes that require urgency. And worse, the criminal network must be able to handle complex information entangled with case-to-case, person-to-person, and case-to-person connections. In this study, we propose a fast inference algorithm for a large-scale criminal network. The network we designed has a unique structure like a sandwich panel, where one side is a network of crime cases and the other side is a network of people such as victims, criminals, witnesses, etc., and the two networks are connected by relationships between the case and its corresponding people. The experimental results on benchmark data showed that the proposed algorithm has fast inference time and competitive performance compared to the existing approaches. After performance validation, the proposed method was applied to the actual crime data provided by the Korean National Police to predict the suspect candidates for several cases.