Enhancing Security in MANETs with Deep Learning-Based Intrusion Detection

Mobile Ad-hoc Networks (MANETs) are dynamic, collaborative networks that maintain reliable connections in a self-organised manner. Intermediate nodes forward data packets between source and destination nodes. However, these nodes are prone to attacks, making intrusion detection systems critical. This paper presents a deep learning algorithm called Graph Neural Network (GNN) to detect possible intrusions in nodes. The GNN is trained with various datasets and tested on the network, due to the mobile nature of MANET nodes, network traffic increases, heightening the risk of unauthorised disruptions. The proposed method is computed using the NS2 simulation tool and shows better resilience to attacks than methods like AODV, ANFIS, and GOA-SVM. The proposed system achieves a Packet Delivery Ratio (PDR) of 90.53%, while AODV, ANFIS, and GOA-SVM achieve 89.13%, 87.69%, and 87.32% respectively for 100 nodes. The method is compared with existing methods in terms of PDR and end-to-end delay during training and testing phases.