ҹɫè

The rapid expansion of the Internet of Things (IoT) has revolutionized various industries, but it has also introduced significant security challenges. Intrusion Detection Systems (IDS) are essential for safeguarding IoT networks, and Machine Learning (ML) has shown promise in enhancing IDS effectiveness. However, existing ML-based methods face critical limitations when addressing high dimensionality and class imbalance in IoT datasets. Current feature selection techniques struggle with adversarial attacks, such as data poisoning, which can degrade model performance, while ensemble learning methods often overlook prediction uncertainty, leading to unreliable detection of minority classes.

This research presents two novel frameworks: a Stability-Enhanced Feature Selection Framework that integrates stability analysis with Recursive Feature Elimination and Cross-Validation (RFECV), ensuring robust feature selection even under adversarial conditions; and an Adaptive Uncertainty-Aware Ensemble Framework that incorporates uncertainty estimation using Monte Carlo Dropout and an active learning mechanism to improve the reliability and accuracy of IDS, particularly for minority classes. The proposed frameworks are evaluated using real-world IoT intrusion detection datasets, and the results demonstrate their effectiveness in enhancing the robustness and reliability of ML-based IDS in IoT networks.

To receive the zoom link, please email Research Training.

For more information, please email the Graduate Research School or phone 07 46311088.