FMCNN: A Factorization Machine Combined Neural Network for Driving Safety Prediction in Vehicular Communication

Abstract

Traffic accidents have globally increased over the past decades, and the driving safety has become an important issue for human society. Despite great progress, the existing driving safety prediction algorithms hardly consider the sophisticated feature interactions between the driving information, traffic information, and driver information. Therefore, in order to solve this problem, a factorization machine combined neural network (FMCNN) is proposed in this paper to predict the driving safety in vehicular communication. In the proposed framework, the factorization machine and the deep neural network are used to learn the effects of low-order and high-order feature interactions from the driving information and the weather information in the pre-training phase, respectively. After the pre-training phase, the high-order feature interactions extracted by the last hidden layer of the deep neural network and the low-order feature interactions trained by the factorization machine are the input of a new deep neural network to predict the driving safety. Varying from the most machine learning algorithms, the proposed algorithm does not require manual extraction of features. It can automatically extract features from the driving information, traffic information, and driver information collected by the vehicle ad hoc network. The experiment results show that the prediction result of the proposed FMCNN is better than DNN and FM in AUC and Logloss.