Vehicle Collision Prediction Model on the Internet of Vehicles

Abstract

AbstractAn active collision prediction model on the Internet of Vehicles is proposed. Through big data calculation on the cloud computing platform, the model predicts whether the vehicles may collide and the time of the collision, so the server actively sends warning signals to the vehicles that may collide. Firstly, the vehicle collision prediction model preprocesses the data set, and then constructs a new feature set through feature engineering. For the imbalance of the data set, which affects predictive results, SMOTE algorithm is proposed to generate new samples. Then, the LightGBM algorithm optimized by Bayesian parameters is used to predict the vehicle collision state. Finally, for the problem of low accuracy in predicting the collision time, the time prediction is transformed into a classification problem, and the Bayesian optimization K-means algorithm is used to predict the vehicle collision time. The experimental results prove that the vehicle collision prediction model proposed in this paper has better results.