Abstract
With the acceleration of urbanization, the spatiotemporal characteristics ofcriminal activities have become increasingly complex. Accurate prediction ofcrime distribution is crucial for optimizing the allocation of police resourcesand preventing crime. This paper proposes LGSTime, a crime spatiotemporalprediction model that integrates Long Short-Term Memory (LSTM), Gated RecurrentUnit (GRU), and the Multi-head Sparse Self-attention mechanism. LSTM and GRUcapture long-term dependencies in crime time series, such as seasonality andperiodicity, through their unique gating mechanisms. The Multi-head SparseSelf-attention mechanism, on the other hand, focuses on both temporal andspatial features of criminal events simultaneously through parallel processingand sparsification techniques, significantly improving computational efficiencyand prediction accuracy. The integrated model leverages the strengths of eachtechnique to better handle complex spatiotemporal data. Experimental findingsdemonstrate that the model attains optimal performance across four real - worldcrime datasets. In comparison to the CNN model, it exhibits performanceenhancements of 2.8\%, 1.9\%, and 1.4\% in the Mean Squared Error (MSE), MeanAbsolute Error (MAE), and Root Mean Squared Error (RMSE) metrics respectively.These results offer a valuable reference for tackling the challenges in crimeprediction.