Machine Learning-Based Risk-Aware Congestion Control Scheme for Minimization of Information Loss in Dense VANET Environment

Abstract

Vehicular Ad-Hoc Network is one of the most growing research fields and has become a promising topic for researchers. Day by day, vehicle density increases, causing very serious issues like road accidents and traffic jams. These issues can be minimized by a periodic exchange of awareness messages among the vehicles. Traffic density increases transmission of awareness messages, which causes channel congestion and degrades safety services, so congestion control schemes must ensure that the channel load should be below a particular threshold, boost the quality of services and minimize information loss and delay in the network. This paper proposes a machine learning-based risk-aware congestion control (MLB-RACC) scheme which is an efficient congestion control scheme based on the modified [Formula: see text] -means machine learning algorithms. MLB-RACC technique processes the input data and helps to improve the quality of services in the automotive industry. It works in three phases: detecting congestion, clustering of vehicles and controlling the broadcasting of awareness messages for decreasing the channel load. MLB-RACC scheme works by grouping the nearest vehicles and the number of groups or clusters is decided on the basis of transmission range/transmission power. The group member selects adaptive transmission rate by looking at the channel busy ratio (CBR). In this technique, message generation rate is controlled at two levels: first is at the clustering level and second is during the checking of CBR. The scheme is implemented through MATLAB, NS2 and the SUMO platform and provides evidence for the minimization of information loss by analyzing the throughput, packet loss and end-to-end delay in comparison to ordinary decentralized congestion control (DCC) technique.