GSNet: Learning Spatial-Temporal Correlations from Geographical and Semantic Aspects for Traffic Accident Risk Forecasting

Traffic accident forecasting is of great importance to urban public safety, emergency treatment, and construction planning. However, it very challenging since traffic accidents are affected by multiple factors, have multi-scale dependencies on both spatial temporal dimensional features. Meanwhile, rare events, which leads the zero-inflated issue. Existing methods cannot deal with all above problems simultaneously. In this paper, we propose a novel model, named GSNet, learn spatial-temporal correlations from geographical semantic aspects for risk forecasting. Spatial-Temporal Geographical Module designed capture among regions, while Semantic proposed model regions. addition, weighted loss function solve Extensive experiments two real-world datasets demonstrate superiority GSNet against state-of-the-art baseline methods.