A Machine Learning based Approach for Protecting Wireless Networks Against DoS Attacks

Two major security threats for wireless networks are physical jamming and virtual jamming. The inherent openness of the wireless channels exposes the network to the physical jamming problem. On the other hand, the virtual carrier-sensing mechanism of IEEE 802.11 based MAC protocols opens up even a less expensive virtual jamming problem. A malicious node can effectively launch Denial of Service (DoS) attacks through virtual jamming inhibiting access to a large portion of a wireless network at the minimum expense of power, resulting in a significant drop in aggregate throughput and traffic carrying capacity of the network. The existing solution(s) can recover to some extent but the attackers are developing new variants of attacks day-by-day. Therefore, novel and more robust mechanisms are needed to combat virtual jamming. In this paper, we propose a novel machine learning based solution that can effectively classify the malicious (i.e., jammer) and non-malicious nodes in order to intelligently ignore any channel allocation requests made by the jammers. Finally, by presenting rigorous simulation results, we show the efficacy of the proposed solution and its superiority over other non machine learning based solutions.