Ensuring the adequacy of neural network models based on the optimization of test volumes

Abstract

© 2019 IOP Publishing Ltd. All rights reserved. Ensuring the accuracy and adequacy of mathematical models based on neural networks is an insufficiently studied area of modeling complex multifactor models based on a limited amount of input data. The article presents the results of studies of randomly organized test samples that complement the training set to the total amount of set of precedents provided for the construction of a mathematical neural network model of the object. It is shown that the adequacy of the neural network model is achieved by agreeing on the size of the set and the expected accuracy of the model on a given set of procedures. The formulation of the optimization problem for adequate neural network models is formulated. The results obtained significantly expand the approaches to increasing the reliability of neural network models.