Hybrid Transformer Deep Learning Framework for Network Intrusion Detection in IoT Environments

Abstract

The rapid expansion of Internet of Things (IoT) technologies has significantly transformed                                           modern communication infrastructures, healthcare systems, industrial automation, transportation systems, and smart city applications. However, the increasing deployment of interconnected IoT devices has also introduced severe cybersecurity challenges due to the presence of vulnerable communication protocols, limited computational resources, and heterogeneous network architectures. Traditional intrusion detection systems (IDS) are insufficient for identifying sophisticated and evolving cyberattacks because they rely heavily on signature-based detection techniques and manually engineered features. Deep learning techniques such as Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks improved attack detection capability by learning hidden traffic patterns automatically. Nevertheless, these models exhibit limitations in learning long-range dependencies and contextual relationships among sequential network traffic data.

To address these limitations, this paper proposes a Hybrid Transformer Deep Learning Framework for Network Intrusion Detection in IoT Environments. The proposed framework integrates CNN, Bidirectional Long Short-Term Memory (BiLSTM), and Transformer architectures to improve attack detection accuracy and reduce false positive rates. CNN is utilized for local feature extraction, BiLSTM captures temporal dependencies, and the Transformer encoder with multi-head self-attention mechanisms learns global contextual traffic relationships. The framework is evaluated using benchmark intrusion detection datasets including UNSW-NB15 and CICIDS2017. Experimental results demonstrate that the proposed model achieves superior performance compared to traditional machine learning and existing deep learning models. The proposed Hybrid Transformer IDS achieves an accuracy of 99.2%, outperforming CNN, LSTM, and Transformer-only approaches in terms of Precision, Recall, and F1-Score. The proposed framework provides an intelligent, scalable, and efficient cybersecurity solution suitable for modern IoT infrastructures.