A centralized and dynamic network congestion classification approach for heterogeneous vehicular networks

Network congestion-related studies consist mainly of two parts: congestion detection and congestion control. Several researchers have proposed different mechanisms to control congestion and used channel loads or other factors to detect congestion. However, the number of studies concerning congestion detection and going beyond into congestion prediction is low. On this basis, we decide to propose a method for congestion prediction using supervised machine learning. In this paper, we propose a Naive Bayesian network congestion warning classification method for Heterogeneous Vehicular Networks (HetVNETs) using simulated data that can be locally applied in a fog device in a HetVNET. In addition, we propose a centralized and dynamic cloud-fog-based architecture for HetVNET. The Naive Bayesian network congestion warning classification method can be applied in this architecture. Support Vector Machine (SVM), K Nearest Neighbor (KNN) and Random Forest classifiers, which are popular methods in classification problems, are considered to generate congestion warning prediction models. Numerical results show that the proposed Naive Bayesian classifier is more reliable and stable and can accurately predict the data flow warning state in HetVNET. Moreover, based on the obtained simulation results, applying the proposed congestion classification approach can improve the network’s performance in terms of the packet loss ratio, average delay and average throughput, especially in the dense vehicular environments of HetVNET.