Advanced Cybersecurity in IIoT: A Machine Learning Perspective on Attack Detection

Abstract

While integrating the Industrial Internet of Things (IIoT) into smart factories massively boosts efficiency, it also opens the door to severe cyberattacks, such as malware and denial-of-service, that can actually disable physical machinery. To protect these vulnerable systems, researchers developed an edge computing-based machine learning framework capable of identifying 14 distinct cyber threats alongside normal network traffic in real time. By processing data on local servers rather than a centralized cloud, the system creates a fast, scalable loop of training, deployment, live prediction, and continuous feedback. When tested on the Edge-IIoTset dataset using a technique called SMOTE to balance the training data, a Random Forest model significantly outperformed Logistic Regression, achieving 85.48% accuracy compared to just 54.55%. Although the Random Forest model proved exceptionally reliable at catching common threats like DoS attacks, it still struggled to identify rarer, stealthier tactics such as man-in-the-middle attacks. Ultimately, this study proves that pairing edge computing with machine learning is a highly effective way to defend industrial networks, even though detecting the most disguised, infrequent cyber threats remains an ongoing challenge for the industry.