STS: A Novel Deep Learning Method for Zero-inflated Crime Prediction

Abstract

Crime prediction is of vital importance to the allocation of police resources and the maintenance of urban safety. Existing methods typically use deep learning to capture the intra-dependencies in spatial and temporal dimensions. However, numerous key challenges remain unsolved, for instance, sparse zero-inflated data due to crimes occurring with low frequency (which can lead to poor performance on real-world datasets), and both intra- and inter-dependencies of criminal patterns across spatial, temporal, and semantic (i.e., categories of crimes) dimensions. In this paper, we propose STS to jointly capture the intra- and inter-dependencies between the crime patterns and the influential factors in three dimensions. Further, we use a multi-task prediction module with a customized loss function to solve the zero-inflated issue. Experiments on two real-world datasets demonstrate the superiority of STS, which outperforms state-of-the-art methods by 26.02% and 18.95% in the mean absolute error and the root mean square error, respectively.