Crime Prediction Using Spatial-Temporal Synchronous Graph Convolutional Networks

Crime prediction is an essential task for improving public safety and optimizing resource allocation. However, accurately forecasting crimes poses significant challenges due to the irregular spatial distributions of crime events and their dynamic temporal patterns. Traditional methods are limited in their ability to capture complex spatial-temporal dependencies. In this paper, we adapt the Spatial-Temporal Synchronous Graph Convolutional Network (STSGCN) to the crime prediction domain, where spatial-temporal relationships and cross-category crime influences are crucial. By leveraging both spatial graph convolutions and temporal convolutions simultaneously, STSGCN can model the intricate dependencies between crime occurrences across different regions and time periods. Our experimental results demonstrate that STSGCN outperforms these models in both crime occurrence prediction and crime frequency forecasting, providing a more robust framework for spatial-temporal crime prediction.