A resource efficient IoT intrusion detection model using hybrid feature selection for edge computing

Area of Science:

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

Background:

• The rapid expansion of Internet of Things (IoT) devices has amplified security vulnerabilities.
• Existing lightweight intrusion detection systems struggle with heterogeneous, imbalanced traffic, often sacrificing detection rates for computational efficiency or vice versa.

Purpose of the Study:

• To introduce BGL-RID (Boruta-Greedy LightGBM Resource-Efficient IoT Detection), a novel framework for resource-efficient IoT attack detection.
• To address the challenges of high detection rates, class imbalance, and computational overhead in IoT security.

Main Methods:

• Implemented a hybrid feature selection pipeline combining Boruta and Greedy Forward Selection (GFS) to identify optimal features.
• Utilized the Synthetic Minority Oversampling Technique (SMOTE) to mitigate class imbalance issues.
• Evaluated performance using accuracy and an efficiency ratio (detection quality vs. resource consumption).

Main Results:

• BGL-RID demonstrated high accuracy across diverse datasets (TONIoT, Raspberry Pi, CICIDS2018, CICIoT2023), achieving up to 99.94%.
• The framework consistently achieved the highest efficiency ratio, indicating a superior balance between detection performance and resource utilization.
• Experimental results confirmed the model's robustness, scalability, and minimal computational/memory requirements.

Conclusions:

• BGL-RID offers a robust and scalable solution for detecting intrusions in IoT environments.
• The framework's resource efficiency makes it highly suitable for deployment on resource-constrained IoT applications.
• BGL-RID effectively balances high detection accuracy with low computational overhead, addressing key limitations of current systems.