Almadhor Ahmad, Altalbe Ali, Bouazzi Imen, Hejaili Abdullah Al, Kryvinska Natalia
Department of Computer Engineering and Networks, College of Computer and Information Sciences, Jouf University, Sakaka, 72388, Saudi Arabia.
Department of Computer Science, Prince Sattam Bin Abdulaziz University, Al-Kharj, 11942, Saudi Arabia.
Sci Rep. 2024 Oct 17;14(1):24322. doi: 10.1038/s41598-024-76016-6.
Due to the rising use of the Internet of Things (IoT), the connectivity of networks increases the risk of Distributed Denial of Service (DDoS) attacks. Decentralized systems commonly used in centralized security systems fail to adequately prevent potential cyber threats in IoT because of the issues of privacy and scaling. The method proposed in this study seeks to remedy these facts by employing Explainable Artificial Intelligence (XAI) together with Federated Deep Neural Networks (FDNNs) to detect and prevent DDoS attacks. Our approach is thus to use federated learning models that are to be trained on distributed and dissimilar sources of data without compromising on the privacy aspect. FDNNs were trained over three rounds with information from three client gadgets incorporating pre-processed datasets of various types of DDoS attacks. Additionally, for feature selection, we integrated XGBoost with SHapley Additive exPlanations (SHAP) to improve model interpretability. The proposed solution can be considered to be quite robust, privacy-preserving, and highly scalable for the detection of DDoS attacks on the IoT network. The results shown on the server side indicate that this approach accurately detects 99.78% of DDoS attacks with a precision rate as high as 99.80%, recall rate (detection rate) going up to 99.74% and F1 score reaching 99.76%. They emphasize that FL-based IDSs are strong enough to cope with cybersecurity challenges in IoT, thus offering hope for securing modern network infrastructures against ever-growing cyber threats.
由于物联网（IoT）的使用日益增加，网络连接性增加了分布式拒绝服务（DDoS）攻击的风险。集中式安全系统中常用的去中心化系统由于隐私和扩展问题，无法充分预防物联网中的潜在网络威胁。本研究提出的方法旨在通过将可解释人工智能（XAI）与联邦深度神经网络（FDNN）结合使用来检测和预防DDoS攻击，从而弥补这些问题。因此，我们的方法是使用联邦学习模型，该模型将在分布式和不同数据源上进行训练，而不会在隐私方面做出妥协。FDNN在三轮训练中使用了来自三个客户端设备的信息，这些信息包含各种类型DDoS攻击的预处理数据集。此外，为了进行特征选择，我们将XGBoost与Shapley值附加解释（SHAP）集成，以提高模型的可解释性。所提出的解决方案对于检测物联网网络上的DDoS攻击而言，可以被认为是相当稳健、保护隐私且高度可扩展的。服务器端显示的结果表明，这种方法能够准确检测99.78%的DDoS攻击，精确率高达99.80%，召回率（检测率）高达99.74%，F1分数达到99.76%。他们强调，基于联邦学习的入侵检测系统足够强大，能够应对物联网中的网络安全挑战，从而为保护现代网络基础设施免受不断增长的网络威胁带来了希望。
