Machine learning for Internet of Things (IoT) device identification: a comparative study.

Area of Science:

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

Background:

• Millions of connected Internet of Things (IoT) devices introduce significant security challenges.
• IoT device identification is crucial for cybersecurity, often using machine learning (ML) on network flows for device fingerprinting.
• Existing studies lack comparative analysis of ML classifiers and feature selection (FS) algorithms for IoT device identification.

Purpose of the Study:

• To comprehensively evaluate the performance of ML classifiers and FS methods for IoT device identification.
• To assess the efficacy of filter- and wrapper-based FS methods across various ML classifiers.
• To introduce and evaluate the Binary Green Wolf Optimizer (BGWO) for IoT device fingerprinting.

Main Methods:

• Comparative performance evaluation of multiple ML classifiers.
• Assessment of filter- and wrapper-based feature selection (FS) methods.
• Implementation and comparison of the Binary Green Wolf Optimizer (BGWO) against traditional ML classifiers using two datasets.

Main Results:

• BGWO achieved significant feature set reduction: 85.11% for dataset 1 and 73.33% for dataset 2.
• BGWO attained high classification accuracies: 98.51% for dataset 1 and 99.8% for dataset 2.
• Wrapper-based FS methods demonstrated effectiveness in reducing feature sets.

Conclusions:

• The Binary Green Wolf Optimizer (BGWO) shows strong capabilities in reducing feature dimensionality and improving classification accuracy for IoT device identification.
• BGWO presents a promising meta-heuristic algorithm for enhancing IoT cybersecurity.
• Wrapper methods are effective for feature reduction in IoT device fingerprinting.