Review of the Application of Lightweight Algorithms for Preventing Internet of Things devices from Distributed Denial of Services Attacks

Abstract

Among the industries that have been revolutionized by the new development of the Internet of Things (IoT) are healthcare, industry, and smart cities but at the same time, it has brought great security threats, notably Distributed Denial of Service (DDoS) attacks. Conventional deep learning intrusion detection systems offer good accuracy in detection, but can be costly in computation and do not fit well in the constrained resource environment of IoT. This review takes a critical look at the newer IoT-based DDoS detection methods with the attention to Federated Learning (FL), Explainable Artificial Intelligence (XAI), and lightweight machine learning (ML) methods. In a comparison of recent literature, it has been found that lightweight ML models including Support Vector Machines (SVM) and K-Nearest Neighbors (KNN) reach a detection accuracy of around 94-96% with low computational and communication overhead, which makes them an appropriate choice in the deployment of edge-based IoT. Conversely, the FL-integrated deep learning methods, such as FL-XAI frameworks and FL-LSTM models, achieve better detection accuracy (99-99.8) and better privacy protection, but pose serious training complexities, communication, and resource constraints on the devices. As a middle ground to scalability, interpretability, and detection accuracy (97-98%), hybrid models like FL-Autoencoders and FL-CNNs exist. Notwithstanding such progress, the majority of investigations are based on simulated data and do not provide the validity of IoT implementation in the real world, which defines a significant research gap. In general, these results indicate that lightweight ML models are the most viable choice in real-time IoT setups, and federated and explainable frameworks are promising the scalability, privacy-aware, and explainable IoT security systems, as long as their computational efficiency and applicability in the real world are further enhanced.