We found a match
Your institution may have rights to this item. Sign in to continue.
- Title
Traffic Accident Risk Prediction for Multi-factor Spatio-temporal Networks.
- Authors
Qingrong Wang; Kai Zhang; Changfeng Zhu; Xiaohong Chen
- Abstract
Predicting the risk of traffic accidents has significant implications for emergency response and urban planning, and is critical in implementing intelligent transportation systems. However, there are some challenges in effectively capturing spatiotemporal correlations in accident- prone regions. Firstly, existing models focus on capturing spatio- temporal features of coarse-grained regions, and need to dynamically integrate predictions for fine-grained regions with those for coarse-grained regions. In addition to weather conditions, environmental factors such as points of interest (POI) and road attributes often influence traffic accidents. Therefore, it is crucial to extract the semantic relation between external factors and accident risk. Thirdly, since traffic accidents are rare events, the training of the model may encounter the problem of zero inflation. To tackle the difficulties above, we present a model for predicting traffic accident risk named Risk- CCNMAGU. Specifically, we design Spatial-channel CNNs and Multi-factor-Attention GCNs to catch spatial features in regions of different granularity. We also implement a dynamic fusion of coarse-grained and fine-grained regions through weighted aggregation. Meanwhile, we construct static, dynamic, and knowledge graph representation views to extract correlations between multiple external factors. We introduce the GRU- Attention module to capture nonlinear temporal correlations to learn temporal features. Additionally, we employ a sample- weighted MSE loss function to alleviate the data sparsity problem. Also, we add introduce to the raw data before feeding it into the model. Finally, we conducted comprehensive experiments on NYC and Chicago datasets, showing that Risk- CCNMAGU outperforms existing models on RMSE, MAE, MAP, and recall metrics.
- Subjects
NEW York (N.Y.); CHICAGO (Ill.); TRAFFIC accidents; INTELLIGENT transportation systems; EMERGENCY management; KNOWLEDGE graphs; KNOWLEDGE representation (Information theory); REPRESENTATIONS of graphs
- Publication
IAENG International Journal of Applied Mathematics, 2023, Vol 53, Issue 4, p1315
- ISSN
1992-9978
- Publication type
Article