A Comparative Study on Machine Learning Algorithms for Congestion Control in VANET

In Vehicular Ad-hoc Networks, vehicles periodically exchange awareness messages for enhanced safety and traffic management. Massively broadcasted messages in a dense environment causes the channel to be congested and further degrades the reliability for safety. As a result, congestion control mechanisms must ensure that the channel load is below a particular threshold. This paper predicts the local density of vehicles and the channel busy ratio using machine learning algorithms such as regression, K-NN, naive Bayes, decision trees, and random forest algorithms without considering beacons. Then, applies a prediction and adaption-based congestion control algorithm to best prediction results. The results of the simulations justify that the P&A-A can adapt the frequency and power of beacons to maintain the load of the network channel under the threshold value (40%).