Generative Adversarial Networks for Energy-Aware IoT Intrusion Detection: Comprehensive Benchmark Analysis of GAN...

Area of Science:

• Cybersecurity
• Machine Learning
• Network Security

Background:

• Internet of Things (IoT) devices face significant security challenges due to resource constraints and imbalanced attack datasets.
• Existing intrusion detection systems struggle with severe class imbalance, impacting the detection of rare but critical attacks.

Purpose of the Study:

• To benchmark five Generative Adversarial Network (GAN) architectures for energy-aware intrusion detection.
• To introduce novel energy-normalized metrics for evaluating GANs in resource-constrained environments.
• To propose an optimized Wasserstein GAN with Gradient Penalty (WGAN-GP) for enhanced attack detection.

Main Methods:

• Evaluated Standard GAN, Progressive GAN (PGAN), Conditional GAN (cGAN), Graph-based GAN (GraphGAN), and Wasserstein GAN with Gradient Penalty (WGAN-GP).
• Introduced energy-normalized metrics: Accuracy-per-Joule (APJ) and F1-per-Joule (F1PJ).
• Developed an optimized WGAN-GP incorporating diversity loss, feature matching, and noise injection for data augmentation.

Main Results:

• Optimized WGAN-GP achieved 99.99% classification accuracy and 100.00% minority class detection on the BoT-IoT dataset.
• WGAN-GP demonstrated superior energy efficiency (Accuracy-per-Joule) compared to Standard GAN.
• Cross-dataset validation showed WGAN-GP achieved 98.40% minority class accuracy, a significant improvement over classical methods.

Conclusions:

• Optimized WGAN-GP offers state-of-the-art performance for energy-aware intrusion detection in IoT.
• Diversity-promoting mechanisms in GAN training enhance both generation quality and classification performance.
• The proposed energy-normalized metrics are crucial for selecting appropriate models in energy-constrained IoT deployments.