Machine Learning Based Zero Trust Architecture for Secure Networking
Saubhagya Munasinghe, Nuwan Piyarathna, Erandana Wijerathne, Upul Jayasinghe, Suneth Namal
Abstract
Traditional perimeter-based security solutions are no longer capable of protecting boundless networking topologies like the Internet of Things (IoT). In contrast, the concept of Zero Trust Architecture (ZTA) can be identified as a prospective solution, since the ZTA is designed not to trust any user, device, or entity even if they are within the network perimeter. However, the adaption of ZTA needs an intelligent system to calculate trust, which is a subjective property that machines are yet to understand in comparison to human beings who take decisions based on intuition, gut feelings, emotional responses, etc. Considering the complexity of trust calculation, this work proposes a novel approach based on machine learning (ML) concepts to realize ZTA in a data networking environment. First, it is necessary to calculate the trust value of a certain entity in order to determine the applicable security policies for ZTA. As the trust is based on various attributes, a large traffic dataset collected from a testbed networking environment is fed into several ML models to derive the best suitable trust attributes. Then these attributes are combined using another set of ML models to evaluate the final trust value of an entity in the testbed environment. Further, this work extends the concepts of online learning to integrate continuous monitoring and fine-tuning of trust metrics as required by the ZTA. Finally, the performance of the system is compared and the results indicate that the proposed model outperforms the traditional perimeter-based security model. Furthermore, the study shows the potential of ML-based network security techniques that can be applied in real-world scenarios to efficiently handle large-volume networks and ensure network security.