Zero-Trust Data Security for Cyber-Physical Systems in 6G-Enabled Smart Environments

Abstract

This paper presents and confirms a zero-trust security design to 6G-enabled cyber-physical systems (CPS), confronting the issues of ultra-low latency, massive connectivity, and tremendous heterogeneity. The framework combines automated data generation, several machine learning models, and overall evaluation strategies to guarantee that the threat is detected in adaptive and robust modes in 6G dynamic environments. An innovative data generation method that can be used in the real world to simulate 6G network traffic and the five most prevalent attack types, such as the distributed denial-of-service (DDoS), man-in-the-middle, spoofing, replay, and zero-day attacks, is presented. With this approach, it was possible to generate a 15 million events dataset of the network with realistic traffic properties and class distribution. The optimization of four detection models LSTM, CNN-LSTM, GRU and Random Forest was done with hyperparameter optimization and five-fold cross-validation. The findings indicate excellent performance of all models, and CNN-LSTM has the best accuracy (98.12) and lowest false positive (0.23), whereas GRU is 23 times faster in the inference process. The weighted ensemble also gave 12.4% more zero-day detection, increasing to 88.7% of unseen threat detection. The framework can detect within 10 ms, achieve a scale of 1,000 nodes, and be resistant against adversarial attacks and concept drift.