Radial basis dual-layer neural network approach for solving nonlinear food web models

The purpose of this study is to solve the food web model by investigating the modified Previte-Hoffman omnivore predator-prey model within food web structures. The model is solved by using a novel stochastic radial basis dual-layered neural network process. The model is an extension of the Lotka-Volterra equations, including Holling Type II functional response to capture the dynamics among three populations: prey, predator, and scavenger. In this dual-layered neural network process, the two hidden layer structure take 18 and 30 neurons, respectively, and a radial basis activation function in both layers. The optimization is accomplished through the Bayesian regularization scheme. A numerical Runge-Kutta solver is used to lessen the mean square error by separating data into training as 70%, testing as 16% and authentication as 14%. The obtained results have been evaluated in approximating solutions for the model by comparing the metrics, error histograms, and regression analyses to data-derived solutions.