Enhancing IIoT Anomaly Detection Using Feature Engineering and Data Balancing

Abstract

The fast development of the Industrial Internet of Things (IIoT) has presented new, unprecedented ac-cess to connectivity and automation of the industrial environment. This connectivity has, however, also served as an important contributor to the increased vulnerability of IIoT networks to advanced cyber threats. The key to the protection of these systems is good anomaly detection; however, the perfor-mance of intrusion detection models is usually constrained by the imbalance of data, noise, and relevance of features in publicly accessible datasets. This paper overcomes these pitfalls by applying a learned fea-ture engineering pipeline together with data balancing mechanisms to enhance the accuracy of detection and resistance. As a case study, the dataset was used to apply the transformation techniques to obtain meaningful features and to eliminate the problem of class imbalance using both manual undersampling and the Synthetic Minority Over-sampling Technique (SMOTE). The models that were considered in eval-uating machine learning were XGBoost, in order to measure how the engineered features and balanced inputs would influence classification performance. The results prove that careful preprocessing, along with smart data balancing, can make the models work significantly better in the context of identifying abnormal behaviours of IIoT networks. This study adds a repeatable protocol of creating more dependa-ble, readable and effective anomaly-detecting systems, especially in multifaceted industrial settings.