摘要：In this paper, we study the application of deep learning in highway road safety prediction. The road safety model is implemented by LSTM and transfer learning. First, using the existing highway road data with known safety levels the training data set is established for supervised learning. We further proposed a novel anisotropic cost function in order to reduce the risks in misclassification. Second, GAN is used to expand the data set in order to provide enough data to train deep model and improve the accuracy of the safety level classification. Third, in order to adapt to different road situations, transfer learning framework is used to share the common knowledge in road safety prediction, which improves the generalization ability of the proposed classification method. Finally, the experimental results show that the proposed method outperformed the conventional method and has a good practical value to improve traffic safety.
