Transfer-Learning-Based Intrusion Detection Framework in IoT Networks

Area of Science:

• Cybersecurity
• Network Security
• Artificial Intelligence

Background:

• Internet of Things (IoT) cyberattacks, particularly zero-day exploits, are rapidly increasing due to inherent network vulnerabilities.
• Traditional Intrusion Detection Systems (IDS) using machine learning (ML) and deep learning (DL) struggle with the scarcity of labeled data in IoT environments.
• Existing DL-based IDSs require large, balanced datasets, which are often unavailable for IoT networks, hindering effective zero-day attack detection.

Purpose of the Study:

• To propose an efficient intrusion detection framework leveraging transfer learning (TL) for detecting zero-day cyberattacks in 5G IoT scenarios.
• To address the challenges of unbalanced and scarce labeled datasets in IoT network security.
• To enhance the detection capabilities of IDSs against novel and evolving cyber threats.

Main Methods:

• Development of an intrusion detection framework utilizing transfer learning (TL), knowledge transfer, and model refinement.
• Implementation of a TL model based on Convolutional Neural Networks (CNNs).
• Creation of three specialized datasets for evaluating the framework's performance in detecting diverse zero-day attacks.

Main Results:

• The proposed TL-based framework demonstrated high accuracy and a low false prediction rate (FPR) in detecting zero-day attacks.
• The framework achieved superior detection rates for various known and zero-day attack families compared to previous DL-based IDSs.
• Experimental validation confirmed the framework's effectiveness in 5G IoT environments with limited labeled data.

Conclusions:

• Transfer learning (TL) is a highly effective technique for improving the detection of cyberattacks, especially zero-day threats, in resource-constrained IoT environments.
• The proposed framework offers a viable solution for enhancing the security of 5G IoT networks facing data scarcity challenges.
• TL-based IDSs present a promising advancement over traditional DL approaches for robust IoT cybersecurity.