通信学报2025，Vol.46Issue(8)：16-30,15.DOI:10.11959/j.issn.1000-436x.2025145
基于数据增强与特征挖掘的异常流量检测方法
Anomaly traffic detection method based on data augmentation and feature mining
摘要
Abstract
To address the limitations of existing anomaly traffic detection methods,such as insufficient recognition accu-racy for minority classes and limited deep feature extraction capabilities,a anomaly traffic detection method based on data augmentation and feature mining was proposed.Firstly,a progressive sampling-based conditional generative adver-sarial network was employed to generate synthetic samples that conform to the distribution of real data,effectively miti-gating learning bias caused by class imbalance.Secondly,a feature correlation matrix was calculated using the Pearson correlation coefficient,transforming traffic data into graph-structured representations to construct a graph dataset.Fi-nally,a multi-layer graph convolutional network with a hierarchical attention mechanism was designed,in which local and global features were hierarchically extracted and fused through a multi-level neighborhood aggregation strategy,sig-nificantly enhancing the model's capability to identify key features.Experimental results demonstrate that the proposed method achieves multi-class classification accuracy of 89.71%and 99.84%on the UNSW-NB15 and CIC-IDS-2017 data-sets,respectively,showcasing its superior detection performance.关键词
异常流量检测/深度学习/生成对抗网络/关联特征挖掘/图神经网络Key words
anomaly traffic detection/deep learning/generative adversarial network/feature correlation mining/graph neural network分类
信息技术与安全科学引用本文复制引用
安义帅,付钰,俞艺涵,刘涛涛..基于数据增强与特征挖掘的异常流量检测方法[J].通信学报,2025,46(8):16-30,15.基金项目
国家自然科学基金资助项目(No.62102422) （No.62102422）
河南省科技攻关基金资助项目(No.242102211070)The National Natural Science Foundation of China(No.62102422),The Key Science and Technology Research Project of Henan Province(No.242102211070) （No.242102211070）
