Robust IoT security using isolation forest and one class SVM algorithms.

Area of Science:

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

Background:

• IoT devices face increasing cyber-attacks due to resource constraints and security protocol limitations.
• Effective anomaly detection is crucial for securing interconnected IoT environments.

Purpose of the Study:

• To develop and evaluate a robust anomaly detection framework for IoT networks.
• To compare the performance of unsupervised machine learning models, specifically Isolation Forest (IF) and One-Class Support Vector Machine (OCSVM).

Main Methods:

• Utilized the TON_IoT dataset for evaluating IF, OCSVM, and a combined scoring approach (CSAD).
• Employed feature importance analysis, cross-validation, and hyperparameter tuning for model reliability.
• Assessed model resilience against adversarial label-flip poisoning attacks and used LIME for interpretability.

Main Results:

• OCSVM outperformed both IF and CSAD in precision, recall, and accuracy.
• The framework demonstrated effective anomaly detection capabilities for IoT environments.
• Lightweight unsupervised algorithms proved suitable for low-resource IoT anomaly detection.

Conclusions:

• Unsupervised machine learning models, particularly OCSVM, offer effective solutions for IoT anomaly detection.
• The proposed framework provides a reliable and interpretable method for enhancing IoT network security.
• Lightweight algorithms are viable for resource-constrained IoT security applications.