Adaptive Voting Mechanism with Artificial Butterfly Algorithm based Feature Selection for IDS in MANET

Abstract

Mobile ad hoc networks (MANETs) have gained more interest from consumers and academics than ever before thanks to the proliferation of wireless networks and the expansion of the benefits and uses of communication networks in general. MANETs are useful in a wide variety of settings since they don't rely on a centralised server or other hardware to relay messages or process data packets. It's one of the primary justifications for implementing MANET in many different domains. However, there are also numerous difficulties that have arisen as a result of these networks' rising popularity, with network security being one of the most crucial. There have been challenges with data transmission and reception due to MANETs' weak regulatory and security frameworks; network infiltration has been identified as one of the most pressing concerns. In MANETs, wireless nodes serve as relays and routers, connecting the source and sink nodes. Accordingly, it is now possible for rogue nodes to penetrate networks and destroy data packets. In order to cope with this issue, modern intrusion detection systems (IDSs) are utilised for remote monitoring of the functioning and actions of nodes present in wireless sensor networks. As well as being able to identify hostile nodes in the network, IDSs can often predict how such nodes will act in the future. In this research work, NSL-KDD dataset is used as an input data. SMOTE and Z-score method are used during pre-processing to remove the irrelevant features and normalize the data. The optimal features are carried out by Artificial Butterfly algorithm and then, finally, ensemble classifiers s. Multilayer Perceptron (MLP), Boosted Regression Trees (BRT) and finally, the adaptive voting mechanism is used to select the best classifier. The results proves that the proposed ensemble model achieved 97.16% of accuracy, where the existing models achieved nearly 95% to 96% of accuracy.