Assessment of the quality of neural network models based on a multifactorial information criterion

Abstract

The paper is devoted to the problem of assessing the quality of machine learning models in the form of neural networks in the
presence of several requirements for the quality of intelligent systems. The aim of this paper is to develop a multifactorial
information criterion that allows choosing a machine learning model in the form of a neural network that best meets the set of
requirements for accuracy and interpretability. This goal is achieved through the development and adaptation of multifactorial
information criteria for evaluating models in the form of neural networks and, in a particular case, three-layer time delay neural
networks used to identify nonlinear dynamic objects. The scientific novelty of the work lies in the development of multifactorial
information criteria for the quality of machine learning models that take into account the accuracy and complexity indicators, which,
unlike existing information criteria, are adapted to the evaluation of models in the form of neural networks. The practical usefulness
of the work lies in the possibility of automatic selection of the simplest machine learning model that provides suitable accuracy when
used in intelligent systems. The practical significance of the obtained results lies in the application of the proposed criteria for
selecting a machine learning model in the form of a time delay neural network for identifying nonlinear dynamic objects, which
allows to increase the accuracy of modeling while ensuring the simplest architecture of the neural network.