Machine Learning Detection Approach Against Trust Manipulation Attack in Trust Management Systems

Abstract

Trust Management Systems (TMS) play a crucial role in establishing and maintaining trust among nodes in various contexts. However, some nodes may adopt suspicious behaviors by strategically alternating between trustworthy and malicious actions to evade detection, an attack called Trust Manipulation Attack (TMA) or on-off attack. This paper addresses the challenge of detecting such suspect behaviors in TMS and proposes a machine learning-based approach for TMA identification. To validate our contribution, we utilize a Vehicular Ad Hoc Network (VANET) dataset generated through a series of simulations. The results obtained from our experiments demonstrate the effectiveness of the proposed approach in accurately identifying nodes with alternating trust patterns, thus effectively detecting suspect behaviors within the TMS. In a machine learning model, the accuracy reached up to 97.78%.