An optimized stacking-based TinyML model for attack detection in IoT networks.

Area of Science:

• Cybersecurity
• Machine Learning
• Internet of Things (IoT)

Background:

• The proliferation of Internet of Things (IoT) devices presents significant security challenges due to increasingly sophisticated attacks.
• Traditional attack detection methods struggle with real-time processing and resource limitations inherent in IoT environments.

Purpose of the Study:

• To propose and evaluate a stacking-based Tiny Machine Learning (TinyML) model for efficient and effective attack detection in IoT networks.
• To address the limitations of traditional methods by offering a solution with low computational overhead.

Main Methods:

• Utilized the ToN-IoT dataset, preprocessed with label encoding, feature selection, and data standardization.
• Implemented a stacking ensemble learning technique combining Decision Tree (DT) and Neural Network (NN) models.
• Evaluated performance using accuracy, precision, recall, F1-score, specificity, and False Positive Rate (FPR).

Main Results:

• The stacked TinyML model achieved a superior accuracy rate of 99.98%.
• Demonstrated efficiency with an average inference latency of 0.12 ms and power consumption of 0.01 mW.
• Outperformed traditional machine learning methods in both detection performance and efficiency.

Conclusions:

• The proposed stacking-based TinyML model offers a highly accurate and efficient solution for IoT network attack detection.
• This approach effectively overcomes the real-time processing and computational overhead challenges faced by traditional methods.
• The model's low resource requirements make it suitable for deployment on resource-constrained IoT devices.