RATE-RAD: A Novel Framework for Robust Anomaly Detection in Network Traffic by Integrating Relational, Adversarial, and Temporal Embedding

Abstract

With increased complexity and volume in interactions within the network, there is a strong need for anomaly detection in network traffic, ensuring that the traditional ML/DL approaches of network traffic monitoring cannot be easily detected. The current approaches cannot capture Relational, Adversarial, and Temporal patterns, leading to limited precision and real-time scalability in anomaly detection. To overcome these problems this study proposes a holistic, multi-method approach by harnessing advanced methods in the detection model, which will incorporate Relational, Adversarial,Temporal insights for Robust Anomaly Detection (RATE-RAD). Starting with GraphSAGE, it captures the real-time detection based on evolving connections from graph-based representations of the network's interactions. SimCLR is a self-supervised contrastive learning framework that produces feature-rich embeddings from raw traffic data, hence contributing to a minimized reliance on labeled data while enhancing the representation. On top of that, Temporal Convolutional Transformers are applied to the sequential traffic data in order to capture long-term dependencies, making recall of anomalies, especially those with temporal nature, easier. CycleGAN can be used for the augmentation of the dataset, injecting synthetic anomalies onto them, hence on the models to make them more robust to the novel threats. The final step consists of a hybrid ensemble model between XGBoost and LSTM that is supported with highly accurate and deducible results.While SEM provides the identification of causal relationships within the identified anomalies. The RATE-RAD model brings about significant improvements such as it increases detection accuracy to approximately 96%, reduces false positives by 15%, and response time becomes faster for analysts because of increased transparency. This model does not only work with regards to the challenges of scalability and accuracy in anomaly detection but also offers actionable insights, rendering it a promising solution for the secure encryption of network environments.