Traffic accident prediction based on deep spatio-temporal analysis

Traffic accidents usually lead to severe human casualties and huge economic losses. Timely accurate prediction of traffic accidents has great potential to protect public safety and reduce economic losses. However, it is a nontrivial endeavor to predict traffic accidents due to the complex causality of traffic accidents with multiple factors, the dynamic interactions of the related factors, and the intrinsic complexity of spatio-temporal traffic data. To overcome these difficulties, this paper provides a novel traffic accident prediction method, namely, STENN, which takes multiple information (Spatial distributions, Temporal dynamics, and External factors) into account, and aggregates these factors by a joint Neural Network structure. To evaluate the proposed method, we collect large-scale real-world data, which include accident records, real-time and citi-wide vehicle speeds, road networks, meteorological condition, and Point-of-Interest (POI) distributions. Experiments on the collected data set demonstrate that STENN is able to predict the traffic accidents in a fine-grained level and the prediction of accuracy outperforms four classical baselines.