High Fidelity Synthetic data with Self Mapped VAE for Intrusion Detection in SDN enabled Heterogeneous 6G Networks

SDN-enabled 6G networks are poised to revolutionize intelligent communication with dynamic control, ultra-low latency, and massive device connectivity. However, this evolution amplifies security challenges, especially for intrusion detection systems that must cope with heterogeneous, rapidly changing traffic and limited labeled threat data. Synthetic data generation emerges as a transformative solution, enabling robust AI-driven security. While Variational Autoencoders (VAEs) are widely used for this task, their reliance on a fixed prior and encoder-decoder KL divergence limits their ability to model complex data distributions effectively. To address this, we propose the Self-Mapped Variational Autoencoder (SM-VAE), which introduces a decoder-driven, data-adaptive latent space without KL-divergence regularization or encoder dependency. Our method enhances the quality and fidelity of synthetic samples, improving model performance for SDN security use cases. Experiments on the InSDN dataset demonstrate superior reconstruction accuracy and learning outcomes, making SM-VAE a powerful tool for data augmentation in securing 6G networks.