Semi-Supervised Learning Based Anomalous Traffic Detection System

In the context of the information explosion of the new era,the number of mobile terminals is growing exponentially and rapidly,which makes the cyber environment increasingly complex.This situation has led to an increasing number of types of cyber attacks,a problem that has caused a great deal of moral and property damage to countries and groups.As a result,the maintenance of network security has become a top priority for scholars’ research.One of the most proven methods of maintaining network security is anomalous traffic detection.Therefore,the focus of this research is to propose the most accurate and efficient method to detect anomalous traffic.In order to solve the problem of difficult feature extraction when the traffic data is unbalanced,and the problem of high dimensional and therefore inefficient feature extraction for anomalous traffic detection.In this paper,we propose a heterogeneous data set fusion method,a normalized flow-based abnormal traffic label propagation method and NF-RSGAN(Normalizing Flow-Residual-Semi-Supervised Generative Adversarial Networks)model.Network)model.This method first fuses a large number of heterogeneous datasets,unifies them and outputs samples in the corresponding format into a customisable reorganised set of data samples to train a more adaptive and better performing model.A proposed normalised flow framework is then used to achieve a more semantically accurate corresponding label for samples with zero known labels or micro-known labels,combined with a priori knowledge.Label propagation is performed,and the data obtained with labels are treated as labelled data,and the massive data crawled in real applications are treated as unlabelled data and fed into the NF-RSGAN model for classification.The method introduces a convolutional self-attention mechanism to improve the residual network,thus deepening the network depth to improve the model performance.The adaptive normalisation flow is also designed to replace the traditional batch normalisation flow,which can find the appropriate normalisation operation for each layer during the training process and improve the performance of the model.Based on the methods proposed above,this paper proposes a semi-supervised learning-based anomalous traffic detection system and implements the concept.The paper presents a general design of the overall architecture of the system,followed by the implementation of a detailed design of each module function,followed by the improvement and implementation of the various improvement schemes proposed above,and finally the validation of the system through simulation experiments.The results of the simulation experiments show that the system is able to detect abnormal traffic in the critical service network and can meet the requirements of high accuracy,low false alarm rate and high efficiency of the abnormal traffic detection system.The anomalous traffic label propagation method based on perceptual normalised flow and the anomalous traffic detection model of NF-SGAN proposed in this paper are of practical value in the current network environment and can meet the needs of various enterprises for system anomalous traffic detection,while safeguarding the privacy and property of citizens,enterprises and society,so this research is of practical significance.