Optimized Intrusion Detection for IoT Networks Using Machine Learning and Feature Selection with RTGBO-ELM Integration
Chandrakanth Reddy Borra, Ramya Vani Rayala, Zabiha Khan, Srinivas Cheekati
Abstract
More than ever before, fraudsters are fixated on the Internet of Things (IoT) because to its astounding rate of growth. The assertion is supported by the cumulative frequency of cyberattacks targeting intermediary communication media and IoT devices. If attacks on the IoT go undiscovered for a long time, they disrupt services severely, which costs money. Additionally, it poses a risk to one's privacy. For IoT-enabled services to be dependable, safe, and lucrative, real-time intrusion detection on IoT devices is crucial. Conventional intrusion detection systems (IDSs) look for common threats using predetermined signatures or criteria, but they could miss more complex or unique attacks. One way to make intrusion detection systems better at detecting attacks is to incorporate ML and DL algorithms into them. Overall, this will strengthen cybersecurity and make it more resilient. Overfitting and the influence of irrelevant characteristics on the discovery of significant patterns are two of the many challenges that ML and DL approaches confront, which can have an influence on the efficacy and performance of the models. Optimising the machine learning models used by intrusion detection systems (IDSs) is necessary to guarantee improved performance and dependability when confronted with novel and unexpected threats. Resolving the issue of overfitting and incorporating feature selection can accomplish this. Here, to present a method for improving intrusion detection in the IoT by preprocessing with machine learning and feature selection. Ring-Toss-Game-Based Optimisation (RTGBO) is used to optimally choose the most relevant features, and the Extreme Learning Machine (ELM) model is used to classify the network traffic. During the experimental presentation analysis, the advocated system shows dependable performance for both simulated and real invasions. It has an average detection accuracy of 97 to 98% for Blackhole, Distributed Denial of Service, Opportunistic Service, Sinkhole, and Wormhole assaults.