Refinement of recommendation trust for accurate detection of dishonest recommenders with on-off attack pattern

Abstract

Recommenders play a significant role in a trust management framework of a mobile ad hoc network, as a node may not have complete knowledge of all its neighbors since it might not have had any interaction with some of its neighbors. Some of the inherent characteristics of mobile ad hoc network like limited radio range, mobility and noise in the channel, contribute to the incomplete knowledge of a node about the forwarding behavior of its neighborhood. Hence the trust management framework has to involve both direct as well as indirect observations for the computation of trust. To ensure accuracy of trust computation, dishonest recommenders have to be filtered out based on a recommender credibility measure termed as recommendation trust. The computation of recommendation trust has to take into consideration the incomplete knowledge of a node about the behavioral characteristics of its neighbors with respect to packet forwarding and providing recommendations. The update of recommendation trust should consider the on-off behavior of the dishonest recommenders which aim to evade detection. The current work presents computation and update modules of recommendation trust which form an enhancement to our earlier works. Simulation results prove the efficiency of the scheme in the presence of dishonest recommenders which exhibit on-off attack pattern.