ORIGINAL RESEARCH article
Front. Phys.
Sec. Social Physics
Volume 13 - 2025 | doi: 10.3389/fphy.2025.1596987
ACSAformer: A Crime Forecasting Model Based on Sparse Attention and Adaptive Graph Convolution
Provisionally accepted
- 1Southwest Jiaotong University, Chengdu, China
- 2Network Security Research Center, Guangxi Police College, Nanning, China
- 3School of Information Technology, Guangxi Police College, Nanning, China
- 4Institute of Software, Chinese Academy of Sciences, Beijing, China
- 5School of Information Technology, Jiangsu Open University, Nanjing, China
Select one of your emails
You have multiple emails registered with Frontiers:
Notify me on publication
Please enter your email address:
If you already have an account, please login
You don't have a Frontiers account ? You can register here
Crime forecasting is vital for urban safety management, as it helps optimize police resource allocation, prevent crimes, and enhance public safety. However, existing supervised learning methods face limitations when processing crime data, such as insufficient spatiotemporal representation capabilities, lack of generalization and robustness, and high computational complexity, all of which restrict forecasting efficiency. To address these challenges, this paper proposes a deep learning-based spatiotemporal sequence forecasting model, ACSAformer,The model integrates a Transformer architecture with adaptive graph convolutional layers and sparse attention mechanisms. The introduction of adaptive graph convolutional layers significantly enhances the model's ability to represent multivariate spatiotemporal sequence, enabling it to capture complex relationships and dynamic correlations among different features, thereby improving generalization and forecasting accuracy. The sparse attention mechanism reduces the number of key pairs that each query needs to attend to, computing similarity scores only for query-key pairs selected according to predefined patterns. This approach lowers the computational complexity from O(L²) to O(L log L), significantly improving efficiency in processing long sequences. We conducted experiments on four datasets from Los Angeles and one dataset from Chicago, all spanning the years 2020 to 2023. The results demonstrate that ACSAformer outperforms traditional spatiotemporal sequence models in terms of forecasting accuracy and robustness.Specifically, for the DS1 dataset, the Mean Squared Error (MSE) is reduced by an average of 17.6%, and the Mean Absolute Error (MAE) is reduced by an average of 9.2%.
Keywords: Crime Spatiotemporal forecasting, Sparse attention, Adaptive Graph Convolutional Layer, Forecasting accuracy, Excellent stability
Received: 20 Mar 2025; Accepted: 26 May 2025.
Copyright: © 2025 Qin, Wei, Gao, Zhu, Qin and Zhang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
* Correspondence: Qian Zhang, School of Information Technology, Jiangsu Open University, Nanjing, China
Disclaimer: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.