Lightweight Deep Learning Method based on Group Convolution: Detecting DDoS Attacks in IoT Environments
Shuanglong Yan, Hongmu Han, Xinhua Dong, Zhigang Xu
Abstract
Distributed denial-of-service (DDoS) attacks remain one of the major security threats in the Internet of Things (IoT) domain. Compared to traditional computing devices, IoT devices typically have more limited computational capabilities and memory resources. To address the resource constraints of IoT devices in DDoS detection, this study proposes a lightweight detection model called DGConv-IDS based on autoencoders and convolutional neural networks. DGConv-IDS adopts a sliding window algorithm to only retain recent data, effectively controlling computational overhead by leveraging the temporal features of DDoS traffic. The model integrates autoencoders and dynamic group convolutional networks into a unified framework, where autoencoders are used for unsupervised feature extraction and dimensionality reduction, and dynamic graph convolutional modules are used for real-time detection of different types of DDoS attacks. For publicly available datasets such as CICIoT2023, we extract multi-dimensional features including timestamps, IP addresses, packet sizes, protocol types, etc. to train the model. Experimental results show that the model achieves high detection accuracy on multiple datasets. Compared with similar deep learning-based methods, the DGConv-IDS model has lower computational costs and better detection performance. In general, the lightweight DDoS detection model proposed in this study is expected to improve the security protection capabilities of IoT devices and provide effective solutions to help IoT systems resist DDoS attacks.