X-FuseRLSTM: A Cross-Domain Explainable Intrusion Detection Framework in IoT Using the Attention-Guided Dual-Path...

Area of Science:

• Cybersecurity and Network Engineering
• Artificial Intelligence and Machine Learning
• Internet of Things (IoT) Security

Background:

• Intrusion detection in heterogeneous and dynamic IoT systems faces challenges due to domain variability and evolving attack tactics.
• Existing methods struggle with cross-domain generalization and providing explainable threat identification.

Purpose of the Study:

• To propose a novel algorithm, X-FuseRLSTM, for effective cross-domain intrusion detection in IoT environments.
• To enhance detection capabilities by incorporating spatial and temporal correlations in network traffic data.
• To provide explainable artificial intelligence (XAI) for model predictions.

Main Methods:

• A dual-path feature fusion framework guided by attention and coupled with a residual LSTM (Long Short-Term Memory) architecture.
• Feature extraction using a deep encoder and sparse transformer, followed by feature fusion and reduction.
• Classification using a deep neural network and RLSTM, with XAI techniques for model explanation.

Main Results:

• X-FuseRLSTM achieved high accuracy across multiple datasets: 99.40% on TON_IoT Network, 99.72% on NSL-KDD, and 97.66% (19-class) and 98.05% (6-class) on CICIoMT 2024.
• The model demonstrated strong domain generalization capabilities.
• The integration of XAI provided explainability for the intrusion detection predictions.

Conclusions:

• X-FuseRLSTM offers a robust solution for intrusion detection in diverse IoT systems.
• The algorithm's effectiveness, domain generalization, and explainability make it suitable for practical IoT security applications.
• The method balances high detection accuracy with computational efficiency.