Efficient Anomaly Detection in Open-RAN Using Conditional Tabular GAN for RIC Applications

The advent of Open Radio Access Networks (Open-RAN) represents a transformative shift in fifth-generation (5G) architecture and future network design, offering significant opportunities and challenges. The inherently open and distributed structure of Open-RAN introduces new threat surfaces, necessitating the development of sophisticated anomaly detection mechanisms. This paper investigates the application of Generative Adversarial Networks (GANs) for near-real-time anomaly detection within the Open-RAN ecosystem, particularly emphasising their integration with RAN Intelligent Controllers (RICs). We propose a Conditional Tabular Generative Adversarial Network (CTGAN) framework, tailored explicitly for RIC applications, to enhance detection efficiency and safeguard data privacy by generating synthetic samples that mimic the original dataset. Through extensive analysis and evaluation, utilising the CICEVSE2024, we assess the performance of CTGAN-generated data compared to the original dataset across five Machine Learning (ML) classifiers. Experimental results demonstrate that the synthetic data generated by the CTGAN achieves optimal performance relative to the original dataset, highlighting the efficacy of this approach for training and testing ML classifiers. Furthermore, this synthetic data generation method ensures data integrity and privacy, as the original sensitive dataset need not be shared with third parties for evaluation purposes; instead, GAN-generated data can serve as a viable proxy while preserving the representational accuracy of the original data. The XGBoost classifier performed the best overall in our work. However, the Naive Bayes (NB) classifier is the highest gainer in our work with +20.69 and +20.18 for Acc., and F1. from its initial performance.