A Densenet-CNN Architecture for Multi-Class DDos Traffic Classification in IIoTs

In this era, industries face extensive cyber-attacks that can cripple production systems and critical operations. Such attacks represent some of the most advanced and nefarious Distributed Denial of Service attacks (DDoS) that exist today. The growing complexity of malicious bot-driven traffic has rendered conventional digital shields and intrusion protection systems largely ineffective. Deep signature systems that can detect malicious and non-malicious actions are exceedingly precise. In balance and consistency, most bots, Gafgyt and Mirai, derive two of the most widely used attack classes, forming a solid base for the defence system. The developed DenseNet-CNN models achieved 88% overall accuracy, with macro-averaged precision at 0.91, recall at 0.87, and F1 at 0.84. The model also memorised benign traffic and all Mirai variants with perfect scores, scoring 1.00 across all metrics. DenseNet CNN still achieved remarkable results, converging quickly with no signs of overfitting and generalising well to a variety of Industrial Internet of Things (IIoT) traffic. This research demonstrates that dense connectivity is a tangible approach for safeguarding IIoT systems.