Lightweight principal component analysis-driven ensemble framework for real-time intrusion detection in industrial IoT networks

Abstract The Industrial Internet of Things (IIoT), as a matter of fact, allows for operational efficiency through integrating real-time data from various resources; however, on the other hand, it also opens new frontiers for cybersecurity risks since the attack surface increases and the environment becomes resource constrained. Conventional intrusion detection systems in most scenarios do not adapt to the changing security requirements of the IIoT. In this research article, a basic lightweight intrusion detection framework that synthesizes Principal Component Analysis (PCA) for reducing dimensions with machine-learning-based ensembles, specifically Naïve Bayes and Random Forest classifiers. Having applied PCA for reducing the original features to 25 and 8 features, correspondingly, the information is well maintained (over 95% of the data variance), with substantially less computational complexity. The proposed system is evaluated on three benchmark datasets, CSE-CIC-IDS2018, CIC-IDS2017, and NSL-KDD, demonstrating robust performance with high detection accuracy, low mean squared error, and sub-millisecond inference latency. The results prove that the new framework maintains high detection performance, increases model generalizability and greatly decreases training and inference times compared with those of full-feature models. These results have justified the framework’s ability to balance accuracy with computational efficiency, offering a scalable and practical solution for real-time intrusion detection in industrial IIoT environments.