Improving Accuracy of Rule Based Collaborative Intrusion Detection System – Security and Performance Trade off in Manet

Abstract

The Intrusion Detection System (IDS) is essential for network security, but its complex environment can result in high false detection rates due to the large number of normal samples. To tackle this issue, an Enhanced Generative Adversarial Network with Bidirectional Long Short-Term Memory and Cross-correlated Convolutional Neural Network (EGAN-BiLSTM-CCNN) has been developed in MANET. This model was deployed in Cluster Heads (CHs) for IDS based on the local information of nodes but faces challenges in capturing global information. Integration issues across diverse clusters hinder its ability to detect coordinated attacks. This study introduces a novel Transfer Learning (TL) mechanism coupled with the EGAN-BiLSTM-CCNN model for IDS. The main objective of this model is to utilize both local and global information of the network based on the TL to enhance the performance of collaborative IDS. First, a cluster-based MANET simulation is established to simulate various attacks such as flooding, black holes, gray holes, and forging attacks. Then, network parameters related to these attacks are collected for each node within clusters and transmitted to respective CHs. CHs share local information to attain a global perspective. By leveraging local and global information, a common latent subspace for various attacks and an optimized representation are discovered based on the TL process, thus generating a training dataset. This dataset is used to train the EGAN-BiLSTM-CCNN model deployed within each CH for intrusion detection, achieving a balance between security and performance in MANETs.