Protocol Anomaly Detection in IIoT

Industrial IoT (IIoT) belongs to the category of Operational Technology (OT) network, which is different from Information Technology (IT) network. The latter can be infused with large amount of computing power whereas the former lacks the support of holding that much computing power since OT network is a resource-constrained area. As a result, a lightweight framework must be built to improve and extend the behavior and performance of the OT network. However, if resource usage is low, the performance will be degraded. A combination of LightGBM and Stochastic Gradient Descent algorithms for anomaly detection is proposed. In the first part, the feature selection process using LightGBM algorithm is discussed and then the model training and anomaly detection methodology using Stochastic Gradient Descent are illustrated. Also, the drawbacks of Gradient Descent algorithm and how the SGD overcomes those drawbacks are discussed. The findings from experiments show that the suggested model generated high throughput and high anomaly detection accuracy.