Exploring Machine Learning Algorithms for Malicious Node Detection Using Cluster-Based Trust Entropy

Abstract

Machine learning has, over the decades, ushered in a dramatic transformation across a range of sectors, including network security. Security experts agree that the potential of machine learning algorithms will be indispensable in detecting all kinds of attacks and maximize accuracy when compared to traditional detection methods. In Wireless ad hoc networks that monitor real-time systems, security remains a concern. Selective forwarding and Denial-of-Service (DoS) are the most common Wireless Sensor Network (WSN) security attacks, resulting in systems making incorrect decisions with negative consequences. Further, the dynamic nature of ad hoc networks creates security issues that hamper effective data communication. While numerous methods have been suggested in the literature to address these issues, there remains a gap for more robust solutions. This paper proposes a novel trust entropy model approach that applies machine learning to significantly improve network security. The proposed cluster-based trust entropy method avoids malicious nodes in routing and re-routing packets effectively along alternate paths. In addition, a new dataset is created from the network simulation results of the proposed method. This dataset serves as the base for applying machine learning algorithms, resulting in exceptionally high detection accuracy. This novel approach not only solves the security concerns, but also raises the standard for accuracy and reliability in Wireless Adhoc Networks.

Keywords

• Back propagation neural networks
• entropy
• geographic routing
• random forest
• support vector machines