articles

The article justifies actuality of application of neural network methods for identification of significant predictors of the transport and logistics infrastructure of regions of the Russian Federation. The condition of the logistics industry of the Russian Federation in comparison with foreign countries has been analyzed. It was concluded that it is necessary to increase the accuracy of estimation of indicators of transport and logistics infrastructure of regions in order to identify their impact on the development of logistics. The problem of the traditional methodology of building a model of transport and logistics infrastructure of regions based on the application of mathematical and econometric analysis lies in the inability of the latter to find and accurately describe the non-obvious dependencies in the data. The expediency of sequential coupling of econometric and neural network research tools has been determined. The two-step procedure of identification of factors influencing the logistics development of the Russian Federation has been tested. As a result, it was possible to select the most significant socio-economic (average per capita income of the population, retail trade turnover, imports of the subjects of the Russian Federation) and infrastructure factors (the share of paved roads, the shipment of goods by public rail, the departure of passengers by public rail, the density of public railway) logistics infrastructure on the basis of an econometric approach. In the second step of the study, a neural network model of the remaining factors was developed based on the development of classification trees and a neural network, acting as a kind of computational filter, which allowed solving the problem of attribution of macroeconomic data and achieving a high level of significance of forecasts. The proposed approach of sequential coupling of econometric methods and neural network modelling has universality and practical importance, therefore it is applicable to the study of a wide range of macroeconomic processes.