Performance Improvements of Covert Timing Channel Detection in the Era of Artificial Intelligence

We are now living in the digital era, where network security is one of the fundamental blocks of communication systems. Digital society, digital economy, and digital communications are now realities in our everyday lives, where sensible data are exchanged through every kind of digital media. Cyber-attacks’ detection methods have hence gained more and more importance to guarantee the system’s security policies, as well to avoid malicious intrusions. This work discusses the application of deep learning (DL) techniques to detect security threats, namely covert timing channels. To evaluate the various deep learning methods that we have explored, we conduct a network attack simulation by manipulating inter-arrival packet delays, effectively creating a covert timing channel. Subsequently, we introduce six customized deep learning frameworks tailored to our specific objectives, designed to differentiate between the presence and absence of a cyber-attack in the data. Our results and comparisons demonstrate the feasibility, robustness, and superiority of DL approaches for improving the network security by detection malicious intrusions.