Identifying malicious nodes in mobile ad hoc networks using a reputation scheme based on reinforcement learning

This paper proposes a reputation mechanism that selects trustworthy nodes for forwarding packets in mobile ad hoc networks (MANETs). The proposed method combines an existing reputation scheme with a reinforcement learning technique called the on-policy Monte Carlo (ONMC) method in the node selection process during the execution of a path search. The objective is to find a decision rule for selecting neighboring nodes which maximizes the long-term average reward. This paper extends a recent work by employing a finite buffer M/M/1/K queuing model to produce packet drops that in turn characterize the reputation values at each node in the MANET. Simulation results show that the proposed scheme can achieve up to 71% and 61% increase in throughput over the existing reputation scheme under both static and dynamic topology cases, respectively.