A Novel Feature-Selection Algorithm in IoT Networks for Intrusion Detection

Area of Science:

• Cybersecurity
• Network Security
• Machine Learning for Security

Background:

• The proliferation of Internet of Things (IoT) devices has amplified cybersecurity risks, making networks vulnerable to malicious activities.
• Intrusion Detection Systems (IDS) are essential for mitigating cyber threats in IoT environments, but their efficiency is challenged by large, complex datasets.
• Feature selection (FS) is critical for optimizing IDS performance by removing irrelevant or redundant data, leading to more effective and timely threat detection.

Purpose of the Study:

• To develop an efficient and rapid feature selection algorithm for enhancing Intrusion Detection Systems (IDS) in the context of the Internet of Things (IoT).
• To address the challenges posed by high-dimensional IDS datasets by implementing a hybrid approach for feature selection.

Main Methods:

• A hybrid wrapper-based feature-selection algorithm, termed CAT-S, was proposed, integrating Cellular Automata (CA) and Tabu Search (TS) with aspiration criteria.
• A Random Forest (RF) ensemble learning classifier was employed to evaluate the fitness of the selected features within the CAT-S framework.
• The proposed CAT-S algorithm was validated using the comprehensive TON_IoT dataset.

Main Results:

• The CAT-S algorithm demonstrated significant improvements in classification accuracy for intrusion detection.
• The method effectively reduced the number of features required for IDS, leading to a more streamlined system.
• A notable decrease in the false positive rate was achieved, enhancing the reliability of the intrusion detection process.

Conclusions:

• The proposed CAT-S algorithm offers a promising solution for developing efficient and accurate Intrusion Detection Systems for IoT networks.
• By optimizing feature selection, CAT-S enhances the practical deployment of cybersecurity measures against evolving cyber threats.
• The study highlights the potential of hybrid metaheuristic approaches combined with ensemble learning for robust network security.