GAT–CNN: Graph Attention Network and Convolutional Neural Network (GAT–CNN) Model for Intrusion Detection in Internet of Things Using BoT-IoT Dataset

Abstract

Abstract

The rapid growth of the Internet of Things (IoT) has caused new problems for keeping networks and devices secure.  The connected devices have become common targets for large-scale attacks. The traditional Intrusion Detection Systems (IDS) and deep learning models often struggle to understand the complex relationships and changing communication patterns that exist in IoT networks. To overcome these issues this study introduces a hybrid Graph Attention and Convolutional Neural Network model for intrusion detection using the BoT-IoT dataset. The model combines the ability of CNN to capture spatial and layered features with the power of GAT to learn relationships between IoT devices through graph-based connections. A complete framework was designed including data preprocessing, graph construction, model development and performance evaluation. In this approach IoT network flows were converted into graph structured data where nodes and edges represent communication similarities that allowing the model to learn device relationships through multi-head attention mechanisms. The hybrid GAT–CNN model achieved an accuracy of 98.9% for binary classification and 99.2% for multi-class classification and performing better than other models such as XGBoost, Ensemble and GCN. The purposed model combines GAT and CNN to improve detection accuracy and consistency. It also shows how IoT networks respond to changing traffic conditions. The findings suggest a practical way to design intrusion detection systems that scale with increasing devices while staying responsive in real time.