Dynamic intrusion detection in AODV-based MANETs using memetic artificial bee colony algorithm

The mobile ad hoc network (MAENT) consists of is a self-configuring network without fixed infrastructure that its topology changes dynamically over time. Due to the inherent characteristics, MANETs are more vulnerable to attacks than wired networks. There are lots of approaches to detect malicious activities in these networks that build a static profile of normal activities and use the profile to identify malicious activities. In MANETs, the use of a static profile is not efficient due to dynamic topology. In this paper, we present a dynamic approach to intrusion detection in AODV-based MANETs, called MemBee, based on a memetic artificial bee colony algorithm. The approach consists of three steps: training, detection and updating. Each node runs a memetic algorithm, called NicheMABC, to generate a set of spherical detectors to cover the non-self space. NicheMABC applies Monte Carlo estimation to prevent generation of unnecessary detectors and a gaussian local search, called GLS, to refine detectors. The spherical detectors are used to discriminate between normal and malicious activities. At specified time intervals, these are updated by one of two methods of partial updating or total updating. We use Monte Carlo estimation to determine when the total updating should be done. We demonstrate the effectiveness of MemBee for detecting several types of routing attacks on AODV-based MANETs simulated using the NS2 simulator. The experimental results show that MemBee can achieve a better tradeoff between detection rate and false alarm rate as compared to other dynamic approaches previously reported in the literature.