PACKET DROP ATTACK DETECTION AND CLASSIFICATION IN MANET USING A NEURAL NETWORK BASED INTRUSION DETECTION SYSTEM

Abstract

In wireless ad hoc networks, communication for end-to-end delivery of packets is achieved cooperatively. The cooperative model ensures that several nodes forming the network coordinate whenever communication is to take place between a sending node and a desired recipient node, which fall out of the sending node’s communication range. This model assumes that an intermediary node will always forward traffic originating from other nodes willingly, other than traffic emanating from the node itself. Conversely, in hostile environments where we find most applications of our ad hoc networks, an always cooperative and submissive behavior on behalf of the other nodes of the network cannot be presumed as the ultimate action undertaken by all the nodes. Misbehaving nodes, which are part of the network, may refuse to pass on traffic to other nodes, for a different many reasons, including to preserve energy or to deliberately degrade performance of the network. Our focus in this thesis is on detecting the presence of malicious nodes that selectively or randomly drop packets intended for other destination nodes, we further classify each packet drop attack, according to its attack type by observing and analyzing how each packet drop attack affect the network characteristics. To effectively detect and classify the misbehaving nodes in MANET, we have developed a system based on the intelligent use of artificial neural networks that makes use of local data collected at each node. Our system has a number of components that work together to achieve the desired objective. The three components are (i) data collection component (ii) data analysis component (iii) detection and classification component. Our three components are integrated together so as to ensure that all malicious nodes present in the network can be detected at high rates with a very low false positive rate. Our technique fares well in comparison to previously proposed methods, with our data analysis component extracting useful metrics and parameters required as input for training our detection and classification engine, which then monitors and evaluates the behavior of each node on the basis of each packet. Using a simulated MANET environment and ANNs modelling we can illustrate that our technique can successfully detect malicious packet droppers as well as classify the several packet dropping attacks at work on the misbehaving nodes.