Machine learning solutions with supervised adaptive neural networks for countermeasure competing strategy of computer virus models

Abstract

The rapid and uncontrolled proliferation of computer viruses in recent years has emerged as a critical challenge, posing significant threats to individual cybersecurity, financial assets, and the stability of industrial systems. This alarming trend underscores the urgent need for advanced and effective strategies to mitigate the spread of such viruses. Motivated by this persistent concern, the current study introduces a robust methodology utilizing supervised adaptive neural networks, backpropagated with the Levenberg-Marquardt method (ANNs-LMM), to model and predict the dynamic interaction between computer viruses and their countermeasures within the nonlinear SIC framework. The nonlinear SIC framework mimics the time-varying dynamic trends of the average values of susceptible, infected, and non-infected internal computers. The referenced solutions for analyzing the dynamic interactions for all three classes of the SIC model are determined for critical significant parameters including the variability in the infection rate of susceptible computers, the acquisition of countermeasures by each susceptible or infected computer, the recovery rate of infected computers, the immunity loss of uninfected computers upon the invalidation of countermeasures, and the distinction between internal and external computers that are not connected to the network by implementing the Adams method. The approximate prediction model for each classification of the nonlinear SIC system is computed using the generated data in ANNs-LMM as responses and predictions for train, validation, and test sampling. Exhaustive simulated scenarios are executed to ascertain the reliability of the presented model for comprehensive advanced resilient threat attacks intended to target key industrial systems. The ANNs-LMM results closely match the observed data, and further assessment using the regression measures, mean squared errors and histogram errors endorses the worth and reliability of the proposed methodology.