A machine learning neural network architecture for the accelerating universe based modified gravity

Abstract

The current investigations present the numerical outputs of the mathematical accelerating universe based modified gravity model (MAUMGM) by designing a computational stochastic structure using the Bayesian regularization neural network. The classification of the mathematical MAUMGM is presented into five different nonlinear classes. A dataset is designed using the explicit Runge-Kutta scheme, which is divided into training as 82% and 9%, 9% for testing and verification. The designed stochastic process for solving the MAUMGM contains log-sigmoid activation function, thirty neurons in the hidden layer, dataset based explicit Runge-Kutta, and Bayesian regularization for the optimization. The correctness of the stochastic solver is perceived by comparing the outcomes along with absolute error 10^-06 to 10^-09. The best training values are reported around 10^-13 to 10^-14, which also signify the solver’s perfection. To authenticate the accuracy, and competence of the solver, some tests have been taken using the parameters of regression, state transition, and error histogram.