Investigating the Effectiveness of Different GNN Models for IoT-Healthcare Systems Botnet Traffic Classification
Phuc Hao, Thanh Liem Tran, Van Dai Pham, Abdelhamied A. Ateya, Tran Duc Le · 2024
Botnet attacks are rapidly increasing and have become a major worry for enterprises globally, requiring quick and effective ways to identify botnet traffic. Healthcare systems are among these networks that suffer from such attacks. The security of healthcare networks is a major issue since criminal networks have advanced and present more complex dangers. In this work, we investigate the effectiveness of various graph neural network (GNN) models, including graph convolutional networks (GCN), graph attention networks (GAT), and GraphSAGE, for Internet of Things (IoT) botnet traffic classification. We consider the IoT to be one of the main enabling technologies of existing and future healthcare systems. The proposed work compares the performance of these models with that of the conventional multilayer perceptron (MLP) algorithm. The classification of botnet traffic is essential for securing IoT devices against various malicious attacks. We employ the publicly available MedBIoT IoT botnet network traffic dataset to evaluate the proposed models and compare their classification performance using multiple evaluation metrics. Results indicate that representing data in a graph form and utilizing GNN models leads to remarkable performance. This finding suggests that GNN models can be a practical approach to classify IoT botnet network traffic.