Deep Learning-based Prediction of Traffic Accident Risk in Vehicular Networks

With the increasing number of vehicle, road traffic safety is in a grim state. Vehicular intelligence can solve this problem well, which is also a problem that needs to be considered in the current industrial Internet of things (IIoT). Therefore, IIoT can get better development with deep learning algorithm. Meanwhile, with the high speed and low delay of 5G communication, 5G-Aided Industrial Internet of Things is rising. This paper proposes a traffic accident risk prediction algorithm based on Deep Convolutional Random Forest Networks (DCRFNs) for edge internet of vehicles. Firstly, the collected real-time traffic data is input into CNN with low complexity. Secondly, after the features are reduced and compressed by the pooling layer, the output of the flatten layer of CNN is transformed into a feature matrix. Finally, according to the evaluation parameters of random forests, the risk of traffic accidents can be predicted. The edge servers select the warning information with high risk of traffic accidents and transmit it to the corresponding vehicle unit. The drivers can reduce the risk of traffic accidents by adjusting their driving status in time. Simulations show that the prediction model DCRFN has higher AreaUnderRoc (AUC), accuracy and lower loss than AdaBoost and CNN.