Deep Temporal Multi-Graph Convolutional Network For Crime Prediction

Urban safety and security play a crucial role in improving life quality of citizen and the sustainable development of urban. In this paper, we propose a Deep Temporal Multi-Graph Convolutional Network (DT-MGCN) model which integrates graph generation component with spatial-temporal component to capture the dependencies between crime and various external factors. More specifically, in the graph generation component, we propose to encode the Euclidean and non-Euclidean correlations among regions into multiple graphs, which will reflect the heterogeneous relationships. The spatial-temporal component which simultaneously employs graph convolutional network (GCN) to capture the spatial patterns and encoder-decoder temporal convolutional network (EDTCN) to describe the temporal features. The experimental results on a real-world crime dataset collected from Chicago demonstrate the effectiveness of the proposed DT-MGCN model, which obtains high accuracy and outperforms the state-of-the-art baselines.