EMLTrust: An enhanced Machine Learning based Reputation System for MANETs

  • Authors:

  • Rehan Akbani;Turgay Korkmaz;G. V. Raju

  • Affiliations:

  • Bioinformatics and Computational Biology, M.D. Anderson, Houston TX, United States;Department of Computer Science, University of Texas at San Antonio, TX, United States;Electrical and Computer Eng., University of Texas at San Antonio, TX, United States

  • Venue:
  • Ad Hoc Networks
  • Year:
  • 2012

Quantified Score

Hi-index 0.00

Visualization

Abstract

Many mission critical networks including MANETs for military communications and disaster relief communications rely on node cooperation. If malicious nodes gain access to such networks they can easily launch attacks, such as spreading viruses or spam, or attacking known vulnerabilities. One way to defend against malicious nodes is to use Reputation Systems (RS) that try to predict future behavior of nodes by observing their past behavior. In this paper, we propose a Machine Learning (ML) based RS that defends against many patterns of attacks. We specifically consider the proposed RS in the context of MANETs. After introducing a basic RS, we propose further enhancements to it to improve its performance and to deal with some of the more challenging aspects of MANETs. For instance, we consider digital signature based mechanisms that do not require trusted third parties, or servers that are always online. Another enhancement uses an algorithm called Fading Memories that allows us to look back at longer histories using fewer features. Finally, we introduce a new technique, called Dynamic Thresholds, to improve accuracies even further. We compare the performance of our RS with another RS found in the literature, called TrustGuard, and perform detailed evaluations against a variety of attacks. The results show that our RS significantly outperforms TrustGuard, even when the proportion of malicious nodes in the network is high. We also show that our scheme has very low bandwidth and computation overhead. In contrast to existing RSs designed to detect specific attacks, ML based RSs can be retrained to detect new attack patterns as well.