Ensemble Filter-based Feature Selection Model for Cyber Attack Detection in Industrial Internet of Things

Abstract

Industrial Internet of Things (IIoT) has been recently observed to be a more vulnerable target for different kinds of cyber-attacks. Most of the attack detection systems that are designed with help of Machine Learning (ML) models suffer from poor accuracy, high false detection rate, and a curse of dimensionality problems. To overcome these issues, an optimal selection of features is required to maximize both detection accuracy and computational efficiency. This paper has proposed an ensemble filter-based feature selection approach in which four well-known filter-based feature selection techniques such as ANOVA, Pearson Correlation Coefficient (PCC), Mutual Information (MI), and Chi-Square (CS) are used to reduce the irrelevant features. Then, in order to achieve two reduced feature sets, features are combined using a suitable design technique (union and intersection operation). To detect cyber-attacks, this reduced feature set is passed to four ML algorithms Decision Tree (DT), Random Forest (RF), XGBoost (XGB), and CaBoost(CTB) classifiers. Using the Edge-IIoT dataset, the effectiveness of the proposed model has been evaluated. Experimental results indicate that the proposed methodology achieved 97.84% and 99.61% accuracy in both the intersection and union feature sets, respectively.