Evaluating MLP and Autoencoder Models for Zero-Day Attack Detection in 6G Networks

Sixth Generation (6G) networks promise a deep integration with Artificial Intelligence (AI), enabling intelligent and efficient communication systems. However, this evolution also introduces new cybersecurity threats, including Zero-Day attacks that exploit previously unknown system vulnerabilities. This paper evaluates two widely adopted models, the Multilayer Perceptron (MLP) and the Autoencoder, for attack detection, with a particular focus on Zero-Day attacks. Using a dataset collected from a real 5G network, the study evaluates each model not only in terms of detection performance but also with respect to computational resource consumption. The results indicate that the MLP model outperforms the Autoencoder in both overall classification accuracy and zeroday attack detection, albeit at the cost of higher computational resource usage.