Poster: Exploring Synthetic Data Generation for Anomaly Detection in the 5G NWDAF Architecture

The transition of paradigm from non-standalone to standalone mode in 5G network creates an opportunity to utilize innovative machine learning and AI-based technology for network traffic analysis. In particular, NWDAF plays a key role in leveraging AI-based models to optimize and enhance the 5G core network functions, including anomaly detection and load balancing. However, it is challenging to ensure the sufficient performance of prediction algorithms under dynamic conditions in NWDAF without guaranteeing the quality and quantity of training data, but only a select group has access to the 5G dataset. To overcome this issue, this paper proposes an approach to generate high-quality synthetic 5G NWDAF data using CTGAN, a specialized generative model that creates synthetic output based on tabular input data. We provide preliminary results of leveraging CTGAN to generate 5G synthetic data and evaluate the synthesizing quality for anomaly detection.