A lightweight intrusion detection method for IoT based on deep learning and dynamic quantization.

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Internet of Things (IoT)

Background:

• IoT devices face significant cybersecurity threats due to extensive network traffic.
• Deep learning models are effective for IoT intrusion detection but are often resource-intensive.
• Limited computing power and storage on IoT devices hinder the deployment of complex detection systems.

Purpose of the Study:

• To introduce a lightweight deep learning model for IoT intrusion detection that addresses resource constraints.
• To enhance the detection performance of IoT security systems against sophisticated cyber-attacks.
• To enable efficient deployment of advanced intrusion detection on resource-limited IoT devices.

Main Methods:

• Developed the DL-BiLSTM model combining Deep Neural Networks (DNNs) and Bidirectional Long Short-Term Memory networks (BiLSTMs) for feature extraction.
• Implemented Incremental Principal Component Analysis (IPCA) for feature dimensionality reduction.
• Utilized dynamic quantization to reduce model complexity and computational burden.

Main Results:

• The DL-BiLSTM model demonstrated superior detection performance compared to traditional deep learning and state-of-the-art methods.
• Experimental results on CIC IDS2017, N-BaIoT, and CICIoT2023 datasets validated the model's effectiveness.
• The model achieved high detection accuracy while maintaining significantly lower model complexity.

Conclusions:

• The DL-BiLSTM model offers an effective and efficient solution for intrusion detection in resource-constrained IoT environments.
• This lightweight approach successfully balances high detection performance with reduced computational requirements.
• The proposed model enhances IoT security by enabling advanced cyber-attack detection on edge devices.