Scholarlink

<p>Since many IoT devices are now in critical infrastructures, it has become harder to manage cybersecurity due to the IDS’ lack of scalability, clarity and adaptability in limited-resource places. It describes an intrusion detection framework that connects curriculum learning with Explainable Artificial Intelligence (XAI) to boost threat detection and explainability in IoT networks. The framework adopts a human-like method to gradually build and train a neural network that incorporates GRUs, LSTM layers and attention mechanisms to properly detect patterns in IoT traffic, no matter how far back in time. A major development is adapting Local Interpretable Model-Agnostic Explanations (LIME) to selectively remove unnecessary features as the model advances, thereby improving both its performance and readability. By restructuring and reducing the model using quantumization and pruning techniques, SqueezeNet can run in real time while maintaining its accuracy, shrinking to just 367 KB. Also, training with Random Forest and XGBoost helps the system to work well in new areas and protects it against attacks. The framework is thoroughly proved on recent and diverse IoT benchmark datasets CIC-IoV-2024, CIC-APT-IIoT-2024 and EDGE-IIoT by achieving 98% accuracy, 99% precision and 97% F1-score. This information confirms that this model reads and handles both traditional and advanced cyber threats with full transparency, effectiveness and is easy to deploy on edge devices. This framework aims to securely and deploy IoT applications on any computer, while detecting intrusions and handling emerging threats.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Luminous defense: XAI-driven adaptive security for critical IoT infrastructures

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月12日

查看全文
  • Faisal S. Alsubaei

摘要

Since many IoT devices are now in critical infrastructures, it has become harder to manage cybersecurity due to the IDS’ lack of scalability, clarity and adaptability in limited-resource places. It describes an intrusion detection framework that connects curriculum learning with Explainable Artificial Intelligence (XAI) to boost threat detection and explainability in IoT networks. The framework adopts a human-like method to gradually build and train a neural network that incorporates GRUs, LSTM layers and attention mechanisms to properly detect patterns in IoT traffic, no matter how far back in time. A major development is adapting Local Interpretable Model-Agnostic Explanations (LIME) to selectively remove unnecessary features as the model advances, thereby improving both its performance and readability. By restructuring and reducing the model using quantumization and pruning techniques, SqueezeNet can run in real time while maintaining its accuracy, shrinking to just 367 KB. Also, training with Random Forest and XGBoost helps the system to work well in new areas and protects it against attacks. The framework is thoroughly proved on recent and diverse IoT benchmark datasets CIC-IoV-2024, CIC-APT-IIoT-2024 and EDGE-IIoT by achieving 98% accuracy, 99% precision and 97% F1-score. This information confirms that this model reads and handles both traditional and advanced cyber threats with full transparency, effectiveness and is easy to deploy on edge devices. This framework aims to securely and deploy IoT applications on any computer, while detecting intrusions and handling emerging threats.