Optimizing Traffic Anomaly Detection with Bertcapsule Model

Traffic anomaly detection is crucial for network security, as it helps prevent potential attacks and maintain network stability. However, the complexity and heterogeneity of traffic data present significant challenges for current methods in feature extraction and anomaly detection. Traditional machine learning techniques rely on human feature extraction, while deep learning approaches, despite their strong characterization capabilities, struggle to capture local traffic patterns, exhibit considerable computational complexity, and tend to focus on a single scenario. To address these challenges, this paper introduces an enhanced capsule network that improves local feature extraction while optimizing the global characterization capability of BERT. The optimized network model effectively balances the learning of traffic features and global anomaly patterns. Additionally, this study conducts experiments across multiple scenarios. The experimental consequence indicates that the method in this paper outperforms the existing methods and can effectively improve the detection accuracy in traffic anomaly detection tasks across various scenarios.