Intrusion detection system with an ensemble DAE and BiLSTM in the fog layer of IoT networks

Abstract

The world is rapidly arriving at the period of the IoT, which connects all types of technology to digital services and provides us with great ease. As the quantity of IoT-connected equipment increases rapidly, there may be a rise in network vulnerabilities, leading to an increase in network threats. Fog computing seems to be a distinctive paradigm that includes the cloud's network's edge, including practical computation and vital infrastructure. As a result of easy access to resources, the fog layer renders the system susceptible to several threats. Tackling these challenges entails detecting intrusions and tracing the route leading to the source of the threat. The objective of this study is to offer a security mechanism and demonstrate how an intrusion detection system can guarantee the integrity of IoT networks. Based on deep learning (DL) approaches, several promising intrusion detection systems (IDSs) have been presented, however, they need time-consuming parameter adjustment in various situations. To address this issue, this study suggests a hybrid Deep Auto Encoder (DAE) and BiLSTM for item installation in the fog due to the need to safeguard essential infrastructure against prompt and efficient identification of malicious threats. Further sparrow search optimization algorithm is proposed for parameter tuning. Utilizing IoT-based data, we assess the effectiveness of our suggested model. The outcome of the experiment obtained by analyzing the suggested IDS utilizing CICIDS2017 and Bot-Iot datasets attested to their supremacy over modern systems that are currently available in terms of precision, accuracy, false alarm rate, and detection rate. To learn more about how well our model works, we added two additional metrics: Cohen's Kappa coefficients and Mathew correlation. The outcomes of our experiments and simulations showed that the suggested approach was stable and reliable across a variety of performance criteria. The experimental outcomes show that the proposed system can effectively describe normal activity inside fog nodes and identify various kinds of attacks.