A Two-Hidden-Layer Shallow Neural Network for Solving the Nonlinear Rumor Spreading Model

Abstract

Motive: The design of a shallow neural network with two hidden layers is presented for solving a mathematical rumor spreading model, which is divided into exposed, stiflers, and ignorant. Method: A structure of shallow neural networks is designed using two-hidden layers taking log- sigmoid and radial basis transfer functions with 14 and 26 neurons. The optimization is performed by the Levenberg-Marquardt backpropagation, which is a robust technique for the nonlinear models. An Euler's scheme is executed to obtain the dataset, which reduces mean square error by separating the data into testing (12%), training (75%), and validation (13%). Results: The exactness of the solver is obtained through the comparison of results, best validation values, and small performances of absolute error. In addition, the accuracy of the designed shallow neural structure is perceived by the histogram error, regression values, and transition of state routines.