Hybrid Optimized Intrusion Detection System Using Auto-Encoder and Extreme Learning Machine for Enhanced Network Security

Abstract

The safety of our networks and the communications to use on a daily basis depends on them. In order to provide safe networks, cybersecurity researchers are highlighting the importance of new, effective intrusion detection systems (IDS). The need for effective IDS grows as attackers continue to create new types of assaults and network sizes continue to grow. By thwarting unauthorised access to the network, IDS also strives to provide availability, integrity, and confidentiality for data sent in networked computers. Machine learning (ML) systems to re used in a number of studies to develop effective IDS; however, researchers began to rely on deep learning (DL) methods when ANNs to reintroduced, which generate features automatically and without human intervention. In order to clean up the input data, this study presents a min-max normalisation approach. Then, a hybrid optimiser called LBOA, which combines Butterfly Optimisation and Lion Optimiser, selects the most relevant characteristics using a deep learning-based Auto-encoder (AE). Lastly, Extreme learning machine (ELM) uses the attributes that to re picked to identify assaults on network traffic. For this aim, to use the UNSW-NB15 dataset, which contains four types of attacks, and the suggested model got an accuracy of around 97%. The suggested method outperforms competing models, according on the comparison findings.