Network Abnormal Traffic Detection Approach Based on Multi-Scaled Deep-CapsNet

It is crucial for network security to be able to detect a variety of threats and unusual network traffic. In order to accomplish aberrant traffic detection, existing detection methods required enormous volumes of data. However, there are rare circumstances in which even a fewshot attack sample can be stopped. The shift in feature attitude will also have an impact on traffic attribute discrimination. However, the standard neural network model is unable to recognize this shift in perspective. As a result, few-shot sample anomalous traffic detection has poor accuracy and efficiency. This research has proposed a strategy for detecting unusual network traffic with a small number of samples. Multi-scale Deep-CapsNet and adversarial-reconstruction (MDAR) made up the system. To begin, the Deep-CapsNet's EM vector clustering is revamped. The prediction is finalized by transforming low-level to high-level features using the attitude transformation matrix. Second, the Deep-CapsNet was improved with the help of a multi-scale convolutional capsule. Finally, a reconstruction classification network (ARCN) that is vulnerable to adversarial input is proposed. Classification of the supervised source data and reconstruction of the unsupervised target data are both completed. In addition, an adversarial training technique that reduced the impact of noise on the reconstruction process is suggested. Using a mixture of MDAR, classification is obtained from a small number of samples (point 4). The results are tested on the ICSX2012 and CICIDS2017 datasets. The experimental outcomes validate the superior training performance of the approach. Additionally, its accuracy is unparalleled in both two- and multi-classification tasks. It is especially useful for real-time few-shot aberrant network traffic detection because to its good anti-noise performance and quick operating time.