Federated Learning-Assisted Coati Deep Learning-Based Model for Intrusion Detection in MANET

Abstract MANET is a set of self-arranged, wirelessly connected nodes. Each mobile ad hoc network node acts as a router to send the packet from the source node to the destination node. MANET nodes’ random movements and decentralized architecture pose security challenges, making them vulnerable to various attacks like node selfishness, network partition, black hole, and DoS due to limited hardware resources. In this paper, a novel Hybrid Intrusion DEtection for MANet (HIDE-MAN) technique has been proposed to detect intrusion like DDoS and MitM attacks in MANET. The proposed HIDE-MAN framework initiates by preprocessing malicious data packets through data cleaning and data transformation resulting in the creation of high-dimensional vectors. The intrusion detection system then makes use of the CO-BiLSTM model, which is based on the actions of Coati and BiLSTM. It categorizes outputs into DDoS attacks, MitM attacks, or the absence of any attacks. Federated learning with GAN networks allows for the aggregation of updates from multiple local models distributed across MANET. Assessment metrics such as accuracy, precision, F1 score, detection rate, recall, and security rate have been utilized to assess the efficacy of the proposed HIDE-MAN method. The comparative analysis shows that the detection rate of the proposed HIDE-MAN is greater by 18.9%, 18.07%, and 4.03% than that of the current KBIDS, WOA-DNN, and MSA-GCNN techniques, respectively.