Detecting attacks on IoT devices with probabilistic Bayesian neural networks and hunger games search optimization approaches

Abstract

Technologies that enable physical objects to communicate with each other are attracting a lot of interest day by day. Internet of Things (IoT) is a communication network that is used to connect with technological devices and systems. The growth of IoT networks has led to malicious targeting of these systems by third parties. To prevent such malicious purposes, network security should be at the forefront. Recently, artificial intelligence-based approaches have been used in security domains. Two accessible datasets were used in this study. The datasets consist of simulated bot IoT network traffic data containing distributed denial-of-service (DDoS) attack types. In the proposed approach, probabilistic Bayesian neural networks and normal Bayesian neural networks are used as the deep learning models. All features of the dataset were trained using two models. Then, the datasets were optimized using hunger games search (HGS) method and the most efficient parameters of the datasets were selected and retrained using deep learning models. In the analysis performed using the proposed approach, an overall accuracy of 100% was obtained for the dataset used in the first experiment and 99.99% for the dataset used in the second experiment. It was also observed that metaheuristic optimization helps the proposed approach to save time in detecting DDoS attacks.