Autoencoding-based self-supervised learning for enhanced representation of network traffic patterns

Network Traffic Monitoring and Analysis (NTMA) is crucial in today’s digital age, especially given the increasing volume of network data and associated cyber threats. Effective NTMA involves analyzing data packets for performance optimization, security, and policy compliance. However, traditional NTMA machine learning methods rely heavily on labeled data, which is costly and scarce. The objective of this work is to develop a self-supervised learning framework, VAE-SSL, that improves network traffic classification by leveraging unlabeled data. Specifically, we aim to enhance network traffic representation learning by designing novel pretext tasks tailored for tabular data, enabling better generalization across classification models. The VAE-SSL employs innovative pretext tasks tailored to tabular data, improving the performance of various classification models, including AdaBoost, SVC, Logistic Regression, and SGD. Our experiments demonstrate that VAE-SSL significantly improves classification metrics compared to traditional supervised learning approaches.