The limitations of unsupervised machine learning for identifying malicious nodes in IoT networks

In today's time, the security in IoT networks interests the scientific community. Indeed, IoT networks are confronted with numerous vulnerabilities, including denial of service, which represents a real threat. The greedy behavior attack is arguably one of the most dangerous and intelligent attacks. Its intelligence lies in the fact that the malicious node executes its attack internally by pretending to be a legitimate node and deliberately falsifying its CSMA-CA parameters. In this paper, we propose a new approach for greedy nodes detection based on an unsupervised machine learning method. In order to evaluate the effectiveness of the proposed method, and to prove the limits of this technique, several attack scenarios were carried out into cooja, and different simulation parameters were taken into account such as the number of packets sent, the energy consumption, and radio status. The detection efficiency of the proposed method is evaluated in two cases, best and worst case. In the first, the detection accuracy is equal to 88.5%, while in the worst it is equal to 86.42%.