Distributed Federated Split Learning Based Intrusion Detection System

作     者：Rasha Almarshdi Etimad Fadel Nahed Alowidi Laila Nassef 

作者机构：Department of Computer ScienceFaculty of Computing and Information TechnologyKing Abdulaziz UniversityJeddah21589Saudi Arabia Department of Computer ScienceFaculty of Computer Science and EngineeringUniversity of HailHail55476Saudi Arabia 

出 版 物：《Intelligent Automation & Soft Computing》 (智能自动化与软计算（英文）)

年 卷 期：2024年第39卷第5期

页      面：949-983页

学科分类：081203[工学-计算机应用技术] 08[工学] 0835[工学-软件工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

主　　题：IDS DFSL DDoS attacks CNN CNN+LSTM 

摘      要：The Internet of Medical Things(IoMT)is one of the critical emerging applications of the Internet of Things(IoT).The huge increases in data generation and transmission across distributed networks make security one of the most important challenges facing IoMT *** Denial of Service(DDoS)attacks impact the availability of services of legitimate *** Detection Systems(IDSs)that are based on Centralized Learning(CL)suffer from high training time and communication *** that are based on distributed learning,such as Federated Learning(FL)or Split Learning(SL),are recently used for intrusion *** preserves data privacy while enabling collaborative model ***,FL suffers from high training time and communication *** the other hand,SL offers advantages in terms of computational resources,but it faces challenges such as communication overhead and potential security vulnerabilities at the split *** Split Learning(FSL)has proposed overcoming the problems of both FL and SL and offering more secure,efficient,and scalable distribution *** paper proposes a novel distributed FSL(DFSL)system to detect DDoS *** proposed DFSL enhances detection accuracy and reduces training time by designing an adaptive aggregation method based on the early stopping ***,the increased number of clients leads to increasing communication *** further propose a Multi-Node Selection(MNS)based Best ChannelBest l2-Norm(BC-BN2)selection scheme to reduce communication *** DL models are used to test the effectiveness of the proposed system,including a Convolutional Neural Network(CNN)and CNN with Long Short-Term Memory(LSTM)on two modern *** performance of the proposed system is compared with three baseline distributed approaches such as FedAvg,Vanilla SL,and SplitFed *** proposed system outperforms the baseline algorithms with an accuracy of 99.70%and 99.87%in CICDDoS20