Advanced Neurocomputational Framework for Assessing Memory-Driven COVID-19 Transmission Dynamics: Integrating Fractional Modeling With Deep Neural Networks

COVID-19 remains a critical global health challenge, necessitating advanced computational methods to accurately predict its transmission dynamics. This study aims to develop a novel neurocomputing procedure based on the Levenberg–Marquardt back propagation neural network for the transmission dynamics of COVID-19 with vaccination and isolation. We investigate the model’s mathematical properties, including global stability, and identify key parameters influencing disease incidence through normalized sensitivity analysis. The numerical solutions of the transmission dynamics of COVID-19 with vaccination are obtained through the reliable stochastic intelligent approach based on the deep neural network (DNN). The model’s accuracy is validated via comparative analysis across all compartments, with performance evaluated through convergence studies, error distribution assessments, and regression analysis. A dataset generated by the modified fractional Adams–Bashforth iterative scheme is used to develop a DNN-based numerical scheme. The DNN architecture consists of an input layer, two hidden layers with 50 and 5 neurons, respectively, and an output layer corresponding to the model state variables. The proposed scheme is then used to minimize the mean square error by dividing the data 15% for both validation and testing, while 70% is applied for training. Seven cases based on the initial conditions with α=0.90,0.92,0.94,0.96,1 are used to analyze the robustness of numerical scheme for the COVID-19 model with different memory effects. The present approach provides solutions that match precisely with benchmark data attaining the best validation results ranging from 10−08 to 10−11, and least absolute error within the range 10−04 to 10−10 across the model equations. In addition to this, the stability analysis and sensitivity results further validate the model’s reliability for real-world applications. This study pioneers applying the DNN utilizing Tanh and ReLU activation functions in the hidden layers to model the dynamics of COVID-19 with various control measures. The proposed hybrid procedure can be efficiently used to solve more complex infectious disease models beyond COVID-19.