Packet Loss Rate Prediction for Vehicular Networks with Regression Methods

Vehicle communication aims to improve the reliability, security and latency performance of vehicle ad hoc network communication. In this study, an heuristic function that calculates the packet loss rate (PLR) in the vehicle network using regression methods is proposed, given different vehicle density, transmission speed, and transmission power values. In order to measure the performance of the regression methods, the transmissions of the connected vehicles under realistic traffic scenarios were obtained by simulations. Simulations are made for different urban and highway conditions by working together with Omnet++, Sumo and Veins environments. The results obtained from these simulations are translated into the PLR dataset. The data set is used as training and test data in various regression algorithms. Numerical results show that Catboost regression method gives the least error between predicted and actual results compared to other regression methods. Thanks to the heuristic we have obtained, given a set of transmission parameters, the PLR can be determined directly. In this way, the system designer chooses among the solutions that provide the given PLR according to the needs, or it becomes possible to reduce the PLR below the target level by re-adjusting the transmission parameters.