Detection of multiple-mix-attack malicious nodes using perceptron-based trust in IoT networks

The Internet of Things (IoT) has experienced a rapid growth in the last few years allowing different Internet-enabled devices to interact with each other in various environments. Due to the distributed nature, IoT networks are vulnerable to various threats especially insider attacks. There is a significant need to detect malicious nodes timely.

Intuitively, large damage would be caused in IoT networks if attackers conduct a set of attacks collaboratively and simultaneously. In this work, we investigate this issue and first formalize a multiple-mix-attack model. Then, we propose an approach called Perceptron Detection (PD), which uses both perceptron and K-means method to compute IoT nodes’ trust values and detect malicious nodes accordingly.

To further improve the detection accuracy, we optimize the route of network and design an enhanced perceptron learning process, named Perceptron Detection with enhancement (PDE). The experimental results demonstrate that PD and PDE can detect malicious nodes with a higher accuracy rate as compared with similar methods, i.e., improving the detection accuracy of malicious nodes by around 20% to 30%.