Predicting Future Incidences of Crime Based on the CNN-Transformer Model

Accurate forecasting of crime numbers is vital for effective urban safety management, enabling law enforcement agencies to optimize resource allocation and implement timely preventive measures. This paper proposes a hybrid crime prediction model that combines Convolutional Neural Networks (CNN) and Transformer architectures to enhance the accuracy of crime forecasts. The CNN module efficiently captures short-term, localized features from crime time series data, while the Transformer component excels at modeling long-term temporal dependencies, allowing the model to predict crime numbers 1 to 3 days in advance. Experimental results, based on real-world crime data from Chicago, demonstrate that the proposed model consistently outperforms several traditional models, including LSTM, TCN, and MLP, across multiple prediction horizons. This performance advantage highlights the model’s effectiveness in both short-term and long-term crime forecasting, providing valuable insights for improving urban safety management and supporting proactive decision-making by city authorities.