Generative Adversarial Network-Based Network Anomaly Detection with Unlabeled Data

In the rapidly evolving landscape of 5G networks, effective anomaly detection is becoming increasingly important to ensure network stability and security. Traditional methods often encounter issues due to the inherent complexity and diversity of network data, coupled with the lack of labeled data. Although there are some feasible artificial intelligence (AI) techniques available, their performance is still limited. To address these challenges, this paper proposes a network anomaly detection scheme. First, the strengths of Autoencoder (AE) and generative adversarial network (GAN) architectures is combined. Through feature transformation, it calculates anomaly scores for network data at specific time points, significantly enhancing detection accuracy. Second, the scheme is deployed in a distributed architecture, thereby improving the robustness and flexibility of network anomaly detection. Finally, the simulation results are provided in two distinct 5G vertical applications to show the performance gains of our proposed scheme.