Fused IGAN-IDS System Leveraging Randomized Tree Classification for Extreme Performance Improvement

Functionality data obtained from real-world data sets exhibit characteristics such as larger sample size and fewer invasive testing methods. Current intrusion detection systems (IDS) have a hard time accurately detecting cyberattacks in IIoT systems, and the massive amounts of information that will drive up the price of training and evaluation. This study utilizes a fused IGAN-IDS approach, which combines IGAN principles with an extraordinarily efficient tree-based choice of features technique, to address these two considerations. To address the issue of imbalanced classes in identifying attackers, one approaches solving it by feeding an insufficient amount of attack sampling into GAN. Then, through understanding the data collection’s features distribution, an alternate sample that is close to the existing attacked sample is output. When dealing with issues involving large quantities of data, the ERT-FS technique is used to pick out important dataset characteristics. Verified on two datasets from real-life, the proposed Fused Improved Generative Adversarial Network for Intrusion Detection Systems (IGAN-IDS) model effectively equalizes the dataset and improves the IDS model’s ability to classify when attacking data is insufficient. The Fused IGAN-IDS (suggested Approach) outperforms all other models assessed, including IGAN-IDS, BSAIDS, AnoGAN, LSTM-GAN, MLP-GAN, and GAN, with an accuracy of $99.8 \%$. Its capacity to handle both regular and attack data successfully is shown by its high categorization capabilities.