A Lightweight Intrusion Detection System for Internet of Things: Clustering and Monte Carlo Cross-Entropy Approach

Area of Science:

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

Background:

• The Internet of Things (IoT) is integral to modern infrastructure but faces significant security challenges due to rapid growth.
• Traditional machine learning-based Intrusion Detection Systems (IDS) struggle with the resource constraints of IoT devices (limited computation and memory).
• Increased vulnerability to cyber attacks necessitates efficient and lightweight security solutions for IoT networks.

Purpose of the Study:

• To propose an enhanced approach for efficient and accurate intrusion detection in resource-constrained IoT environments.
• To overcome the limitations of existing IDS solutions in handling IoT device constraints.
• To develop a lightweight, efficient, and high-accuracy IoT-based IDS.

Main Methods:

• Implemented a recursive clustering method for data condensation using compactness and entropy-driven sampling to create a representative data subset.
• Utilized a Monte Carlo Cross-Entropy approach combined with feature stability metrics for selecting the most relevant and stable features.
• Evaluated the proposed approach on N-BaIoT, Edge-IIoTset, and CICIoT2023 datasets from real IoT devices.

Main Results:

• Achieved comparable classification accuracy to existing methods on multiple IoT datasets.
• Significantly reduced training time by 45× and testing time by 15×.
• Lowered memory usage by 18×, demonstrating a highly efficient IDS.

Conclusions:

• The proposed approach effectively enhances intrusion detection efficiency in IoT networks.
• The method provides a lightweight and accurate IDS suitable for resource-constrained IoT devices.
• This research offers a viable solution to bolster IoT security against cyber attacks.