Deep Attentional Factorization Machines Learning Approach for Driving Safety Risk Prediction
Jun Zhang, Zhongcheng Wu, Li Fang, Wenjing Li, Tingting Ren, Wei Li, Jie Chen · Journal of Physics Conference Series · 2021
Abstract The data of Internet of Vehicles (IoV) can be used to evaluate the driving safety risk of auto insurance policyholder and provide technical means for Usage Based Insurance (UBI). There are many types of IoV data, such as continuous or ordinal, categorical, binary etc., which contain highly sparse and dimensional features after One-Hot processing, thus they learning the interaction between critical features and training predictive model difficult. Furthermore, some of the available data have been desensitized, so it is impossible to perform feature engineering based on experience. We propose an end-to-end deep learning framework named Deep Attentional Factorization Machine (DeepAFM), which combines the power of attentional factorization machine with deep learning for feature learning in a new neural network architecture. Compared with existing deep learning models, our approach can learn the weighted interactions between various features effectively by introducing the structure of feature fields without feature engineering. Experimental results showed that our model yields excellent results in real-world data.