A graphics processing unit assisted CNN-GRU framework for the intrusion detection mechanism in the industrial internet of things

Abstract In recent years, there has been a remarkable surge in cyber-attacks and security breaches targeting the Internet of Things (IoT) and Industrial IoT (IIoTs). In today’s computing world, safeguarding the IIoT network and its underlying architecture has become a top priority. Given the nature of sensor networks and the communication environment, it is often difficult to identify and thwart undesired network activity. In today’s networks, network intrusion detection systems are being implemented to detect various harmful activities that may be performed. To begin with, an innovative network intrusion detection (ID) model has been addressed to enhance the problems posed by existing intrusion detection models. In this work, an advanced hybrid structure model that integrates Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) is proposed for identifying botnets and cyber-attacks in industrial sensors and camera networks. In addition, this architecture addresses the weaknesses of traditional intrusion detection systems, such as limited computational capacity and the requirement for highly expansive computing platforms to deploy real-time models. Consequently, a framework for multiclass intrusion is developed and tested in the NVIDIA Jetson Nano Graphic Processing Unit (GPU) computer. The performance analysis of the developed model was conducted using various datasets and compared with CNN and GRU-assisted networks. The CNN-GRU framework demonstrates the best performance for accuracy, at 89.46%; precision, at 95.67%; recall, at 90.82%; F1 score, at 88.07%; weighted average (0.96); and macro average (0.94). These results also emphasized the importance of a hybrid software model combined with GPU hardware to ensure maximum intrusion detection performance for IIoT networks, including smart cities and smart homes.