Explainable AI Approach Enhances Cybersecurity in Internet of Things

Strategic Insights -

Researchers from Shiraz University of Technology have developed an explainable AI approach to intrusion detection in the Internet of Things (IoT). This method uses a lightweight one-dimensional convolutional neural network (1D-CNN) that allows for the interpretation of results and identifies the most important features in the model. The proposed method has shown high accuracy and precision in predicting attacks, with an F1-score of 0.9947.

The research aims to address the issue of interpretability in Machine Learning (ML) and Deep Learning (DL) techniques used in Intrusion Detection Systems (IDSs). The proposed method uses SHapley Additive exPlanations (SHAP) technique for feature selection and Agnostic methods for local and global interpretations of the results. The experimental results using the TON-IoT dataset showed accuracy, precision, recall, and F1-score criteria to 0.995, 0.9949, 0.9947, and 0.9947, respectively.

This research has significant implications for improving the security of IoT networks, which are increasingly vulnerable to cyber threats. The proposed method provides transparency and interpretability of IDS judgments, allowing cybersecurity personnel to gain a better understanding of the network's vulnerabilities.

Key Takeaways:

  • The proposed hybrid model combines a lightweight 1D-CNN with the interpretation ability of the results, making it suitable for resource-constrained IoT devices.
  • SHAP technique is used for feature selection to detect the most important features in the model, which can be used to redesign the model and reduce computation complexity.
  • Agnostic methods are employed for local and global interpretations of the results, providing transparency and explainability of the IDS judgments.
  • The proposed method has shown high accuracy and precision in predicting attacks, with an F1-score of 0.9947.
  • The TON-IoT dataset has been used to evaluate the proposed method, which consists of a collection of IoT network traffic data.
  • The research aims to address the issue of interpretability in ML and DL techniques used in IDSs, which is essential for improving the security of IoT networks.
  • Fatemeh Ebrahimi, Reza Javidan, Reza Akbari, and Yasin Hosseini are the authors of this research, which has been published in the journal Cybersecurity.

Statistics:

  • Accuracy: 0.995
  • Precision: 0.9949
  • Recall: 0.9947
  • F1-score: 0.9947

Sources:

  • Intrusion detection in the internet of things using convolutional neural networks: an explainable AI approach. Cybersecurity, 2025,8(1):1-22.
  • Cybersecurity - https://cybersecurity.springeropen.com/
  • doi-org.sdpl.idm.oclc.org/10.1186/s42400-025-00369-2