RISK PREDICTION OF THEFT CRIMES IN URBAN COMMUNITIES: AN INTEGRATED MODEL OF LSTM AND ST-GCN

Urbanization has transformed urban communities but also introduced challenges in managing crime and ensuring public safety. For prevention and control to be effective, a criminal risk prediction system is necessary. This paper presents a model that integrates Long Short-Term Memory Networks (LSTM) and Spatial-Temporal Graph Convolutional Networks (ST-GCN) to identify high-risk areas in cities. Using topological maps and crime data from Chicago, the model extracts spatial-temporal and temporal features to analyze theft patterns. Experimental results demonstrate its ability to predict crime occurrences within specific timeframes, offering a valuable tool for urban safety planning. The objectives of this study are (1) to develop an integrated LSTM + ST-GCN model for theft crime risk forecasting in urban communities, (2) to validate the model on real-world Chicago crime data, and (3) to benchmark its performance against standard baselines. Unlike existing methods that focus solely on spatial or temporal patterns, our hybrid approach fuses both dimensions for near real-time risk prediction, demonstrating up to 15% lower RMSE compared to standalone LSTM models