Optimizing Feature Selection for IIoT Security: A Sequential Machine Learning Strategy

Abstract

IIoT attacks have the potential to stop industrial processes, which can result in lost production, broken equipment, and large financial losses. IIoT systems frequently manage sensitive data, including operational, proprietary, and employee personal data. Attacks may result in information theft, misuse, and data breaches. To depict the behaviors and interactions of IIoT dev ices, extract pertinent elements from the raw data. features may consist of command sequences, resource use measurements, and device communication patterns. Most of the existing systems uses tree-based techniques or embedded techniques like LASSO to further hone the feature set in accordance with model performance.  It may not be ideal if all correlated features are significant because LASSO has a tendency to randomly choose one feature from a set of highly correlated data and ignore the others. The proposed model designs a sequence of ml models namely Random Forest followed by Integrated KNN and logistic regression as feature selector because this order guarantees that the features are assessed using several viewpoints: ensemble techniques (Random Forest), non-parametric methods based on distance (KNN), and linear models (Logistic Regression). The balancing of complexity, performance, and interpretability can be accomplished by using these techniques. A more robust feature selection procedure results from each phase that lessens the drawbacks of the one before it.