A cyber-attack detection model for IoT-IDS utilizing an Elman Recurrent Neural Network
Abstract
The Internet of Things (IoT) connects a wide range of physical devices through the Internet and is widely adopted in many fields such as transportation, defense, healthcare, agriculture, and more. Its growing popularity originates from its ability to address real-time challenges efficiently. However, the use of diverse transmission and communication protocols has introduced significant security risks, making traditional security methods, such as rule-based or signature-based systems, less effective. To ensure robust network protection, it is crucial to analyse traffic behaviour and detect cyber threats. In response, this research proposes a DL (deep learning)-based IDS (Intrusion Detection System) that leverages the ERNN (Elman Recurrent Neural Network) to detect anomalies in network traffic. The BBOA (Binary Butterfly Optimization Algorithm) is utilized for feature selection. The model is evaluated under two experimental settings: (i) binary classification (benign vs. attack) and (ii) multiclass classification that utilizes the same ERNN architecture to classify different attack categories. Two datasets are used to evaluate the model: CIC-DDoS-2019 and CIC-IoT-2023. Experimental results show that the CIC-DDoS-2019 dataset provides the best performance, achieving an accuracy of 99.12%, a detection rate of 98.95%, a precision of 98.76%, an F1-score of 98.85%, and a specificity of 99.22%. The CIC-IoT-2023 dataset also yields strong results, attaining an accuracy rate of 98.87%, a detection rate of 98.72%, a precision of 98.41%, an F1-score of 98.56%, and a specificity of 98.94%. BBOA for significant feature dimensionality reduction with ERNN to achieve high detection accuracy and reduced computational requirements for IoT environments with limited resources.