Bridging digital and physical realms: cross-evaluating intrusion detection models in cyber-physical systems

Abstract

The application of digital models in industries is recently being used as a means to enhance the operational efficiency of cyber-physical systems (CPSs). However, this comes with challenges bordering mostly around the detection of cyber-attacks using replication as a means. This study considers a synthetic digital object dataset and its equivalent physical object for intrusion detection using principal component analysis (PCA) and an autoencoder for dimensionality reduction. Application of random forest and XGBoost machine learning on both objects is applied to compare the performance on the two objects. To fill a methodological gap, the study implemented the developed model on the digital object dataset and its physical object equivalent to assess intrusion detection capability. The study established that the application of the developed random forest model on the use case digital object dataset varies slightly from its physical equivalent as assessed using standard performance metrics, though revealing very good performance for both objects, which validates the digital object and its applicability in the real world.