Optimized ensemble machine learning model for cyberattack classification in industrial IoT.

Area of Science:

• Cybersecurity
• Machine Learning
• Industrial Control Systems (ICS)
• Internet of Things (IoT)

Background:

• Industrial control systems (ICS) and IoT devices face increasing cyber threats, particularly in critical sectors like oil and gas.
• Existing machine learning (ML) methods for cyberattack detection often lack computational efficiency and rely on binary classification.

Purpose of the Study:

• To propose and evaluate optimized stacked ensemble models for enhanced cyberattack detection in ICS and IoT environments.
• To reduce computational overhead while improving detection accuracy for sophisticated cyber threats.

Main Methods:

• Developed two optimized stacked ensemble models integrating diverse base learners (Logistic Regression, Extra Tree Classifier, XGBoost, LGBM, RFC).
• Selected models to address challenges in security datasets like class imbalance, noise, and complex attack patterns.
• Evaluated models on their ability to leverage different decision boundaries and learning mechanisms for improved detection.

Main Results:

• The Stacked Ensemble_2 model achieved 97% accuracy with a computation time of 54 minutes.
• Stacked Ensemble_2 outperformed Stacked Ensemble_1 and reached 100% accuracy with a 99% AUROC on the CICIDS 2017 dataset.

Conclusions:

• The proposed Stacked Ensemble_2 model offers a scalable, real-time solution for securing ICS and IoT environments.
• Demonstrated significant advancements over traditional methods in accuracy and efficiency for detecting cyber threats in critical infrastructures.