An ensemble of fuzzy soft expert set with deep learning on attack detection for secure industrial cyber-physical systems

The most effective tool for forming uncertainty in problems of decision-making is neutrosophic set (NS) and its additions, like interval NS (INS), complex NS (CNS), and interval complex NS (ICNS). An effective tool for signifying vagueness and uncertainty in the decision process is NS, which is more generalization of a classical set, fuzzy set (FS), and intuitionistic fuzzy set (IFS) by including 3 degrees of falsehood, truth, and indeterminacy of a definite statement. A cyber-physical system (CPS) combines numerous connected physical methods, networking units, and computing resources. Also, it observes the applications and processes of the computational techniques. Interconnection of the cyber and physical world starts threatening security tasks, particularly with the enlarging intricacy of communication systems. Despite the struggles to contest these tasks, it is very complex to identify and examine cyber-physical threats in a compound CPS. Many researchers have implemented machine learning (ML) and deep learning (DL)-based methods for analyzing cyber-physical security methods. This study develops an Enhanced Single Valued Model using the Heuristic Search on Attack Detection (ESVM-HSAD) model in Industrial CPS. The foremost intention of the ESVM-HSAD technique is to concentrate on the recognition and classification of cyberattacks in CPS. The ESVM-HSAD method utilizes the Grey Wolf Optimizer (GWO) as a feature selection (FS) technique to select an optimum set of features. For attack recognition, the ESVM-HSAD methodology utilizes a single-valued fuzzy soft expert set (SVSFES) method and an ensemble of two DL classifiers, such as GRU and Convolutional auto-encoder (CAE). Finally, the recognition outcomes of the ensemble model are enhanced by utilizing the Whale Optimization Algorithm (WOA). Many experiments were performed to detect the improved performance of the ESVM-HSAD approach. The comparative study of the ESVM-HSAD approach exhibited a superior accuracy value of 99.01 % when equated to other methods.