A Secure Multiphase Detection And Analysis of Malicious Node Behaviour in MANETs

Abstract

MANETs, or mobile ad hoc networks, are said to be decentralized with fewer infrastructures, with each device acting as a forwarding router and sending packets as routers do in the network. Whenever there is a need for data transfer, the node will identify its neighboring nodes and the data will transfer to the destination through them. Due to its major characteristics like its dynamic topology and insecure medium, it is prone to various categories of attacks. newlineThe most prevalent well-accepted and best routing mechanism which are having large usage in MANET is an AODV, an on-demand, routing. In many situations, an AODV standard is influenced by many attacks wherein the node intends to send fictitious routing reply information stating to the respective sender that it is having a very direct path to the receiver, but rather the intention is to drop the packet. These categories of attacks are called black-hole attacks. newlineIn this report, MANET s performance is investigated by considering single and also multiple Blackhole attacker nodes. And further, the intrusion detection system for identifying the blackhole attack is analyzed. The proposed methodologies are implemented in NS-2.35 and NS-3.25 platforms and the outcomes are discussed by considering various network configurations like average goodput, delay from end to end, the overall count of lost packets, and jitter. Results also demonstrated that the testing is carried out by using both UDP packets and TCP packets. newlineFurther, the various types of attacks are predicted by leveraging several machine learning methods and algorithms, and the capability of the learning approaches like accuracy , precision, and recall are analyzed with its results. As a result, the detection rates of Black hole attacks are discussed by changing the network settings like the count of attacker nodes, number of nodes, mobility range, energy consumption, and bandwidth. Different types of attacks are classified and learning model parameters are compared against the traditional machine learning ap