东南大学学报（自然科学版）2025，Vol.55Issue(5)：1486-1492,7.DOI:10.3969/j.issn.1001-0505.2025.05.031
基于Layer-wised个性化联邦学习的丢包加密流量分类
Packet loss encrypted traffic classification based on layer-wised personalized federated learning
摘要
Abstract
In the open client-edge-cloud architecture,the performance limitations of data collection and analy-sis devices often lead to packet loss in traffic samples collected at high-throughput traffic gateways,which ad-versely affects the training and application of network traffic classification models.To address this issue,a layer-wised personalized federated learning model training scheme was proposed.This scheme established dis-tributed nodes in traffic sample data under different packet loss rates,collaboratively trained classification mod-els,and dynamically adjusted parameter weights to achieve precise identification of encrypted traffic samples under different packet loss rates.The experimental results show that this method successfully completes the ap-plication classification task for encrypted traffic with a packet loss rate of no more than 20%.In scenarios with extremely uneven sample distribution and low-quality data,the classification accuracy exceeds 88%.The ef-fectiveness of the proposed method in high packet loss environments is validated,providing a new solution for the classification of encrypted traffic.关键词
加密流量分类/联邦学习/个性化机器学习/网络测量Key words
encrypted traffic classification/federated learning/personalized machine learning/network mea-surement分类
信息技术与安全科学引用本文复制引用
秦天,程光,卫亦辰,何濛..基于Layer-wised个性化联邦学习的丢包加密流量分类[J].东南大学学报（自然科学版）,2025,55(5):1486-1492,7.基金项目
国家自然科学基金联合基金资助项目(U22B2025) （U22B2025）
国家自然科学基金面上基金资助项目(62172093) （62172093）
国家自然科学基金青年基金资助项目(62202097) （62202097）
中国博士后科学基金资助项目(2024T170143). （2024T170143）
