A Self-Supervised Learning Approach for Zero-Day Attack Detection in Network Traffic Analysis

This paper presents a self-monitoring approach for zero-day stop detection in network traffic analysis. The plan integrates self-monitoring with an integrated network to identify past attack patterns without relying on technical data show. A novel pre-training mechanism is developed to learn robust feature representations from unlabeled network traffic, enabling effective detection of zero-day attacks. The architecture employs a hierarchical feature extraction approach combined with an adaptive detection mechanism that continuously updates to address concept drift in network traffic patterns. The evaluation has been carried out by various benchmarks, including CICIDS2017 and TON-IoT, showing the effectiveness of the plan. The system achieves a detection rate of 97.8% for zero-day attacks while maintaining a false positive rate of 1.2%, outperforming existing methods. The self-monitoring function of learning reduces the dependence on registered data while maintaining high accuracy. The system demonstrates high performance in various tasks, achieving up to 100,000 flows per second with minimal workload. The evaluation results confirmed the good use in the global network environment, especially in situations where traditional educational monitoring is limited due to the lack of training materials.