Hybrid honey badger-world cup algorithm-based deep learning for malicious intrusion detection in industrial control systems

Abstract

The security analysis has become the hotspot concern in Industrial Control Systems (ICSs) that grabs the research attention in today’s era. Owing to the rapid admittance of the Industrial Internet of Things (IIoT), the superimposition of fog and cloud computing plays a pivotal role for rectifying such issues as fewer computation resources and more latency in the network. In addition to this, security issue comes first while developing the ICS in the sector of IIoT. In general, ICS is applied for various processes, such as electric utilization and water supply management, that are related to every individual’s life. Though it is interlinked with the Internet, it can easily expose to different threats. To defend the model, Intrusion Detection Systems (IDS) are employed. In signature-based detection, anomaly detection is used to find out the unknown attacks in the network. Yet, the ICS is structured with many software and hardware tools. It also seems in the openness nature of the industrial environment that, it becomes vulnerable to various attacks. Hence, detecting unknown attacks is a complex task in the openness nature of the industrial environment. It causes failure to protect such systems from malicious entities that could result in more complications for humans. Hence, the detection of malicious activities is essential. In order to provide an efficient model, a new hybrid deep learning is proposed in ICS. Initially, the original data is to be collected based on the ICS industry. Secondly, the gathered data is preprocessed to clean the artifacts and clean the data. Thirdly, the optimal features are chosen directly from the gathered data with the help of a hybrid algorithm called the Hybrid Honey Badger-World Cup Algorithm (HHB-WCA). The chosen optimal features are fed to the hybrid deep learning termed as Autoencoder-Bi-directional Long Short-Term Memory (A-Bi-LSTM), where the encoded features from Autoencoder are extracted and encoded features are subjected to the Bi-LSTM classifier. Finally, the simulation outcome has shown that the developed method is compared with the other traditional algorithms to detect the unknown classes in the network.