OPTIMIZED IRDN-IBDA: AN OPTIMAL FEATURE SELECTION AND RECOGNITION FOR NETWORK INTRUSION DETECTION SYSTEM

The demand for an efficient intrusion detection system has grown as attackers continue to create new attacks and network sizes expand. Recently, many techniques have been released for network intrusion detection systems (NIDSs). However, new threats are constantly developing and outside existing systems reach. The intrusion detection algorithms high error rate, significant dimensionality, false alarm rate, redundancy, meaningless data, and false negative rate now in use are only a few of the many issues with them. Given its exceptional performance in various detection and recognition tasks, we present a novel and efficient deep learning-based NIDS in this research. Initially in preprocessing data encoding and normalization are performed using raw input data. After preprocessing, the pre-processed data are fed into the feature extraction phase. The features are extracted by utilizing the SE-ResNeXt-101 approach. Then, the essential features are selected with the help of an Improved Binary Dandelion Algorithm (IBDA). The presented novel Improved Residual Dense Network (IRDN) is employed to identify attacks which enhance security and privacy inside the network framework. The lyrebird optimization technique is used to tune further the hyperparameters derived from the IRDN approach to increase performance. The Modified Generator GAN (MG-GAN) algorithm also solves the data imbalance issue. The research shows that the suggested technique outperforms current NIDS methods regarding assessment metrics. Additionally, this method is more suitable for complicated detection of network intrusion requirements.