Use of Neuro-Net Modeling for Forecasting Key Finance Figures at Trade Enterprises

Abstract

The article studies methods of raising efficiency of managing the trade chain ‘M. Video – Eldorado’ on the basis of introducing neuro-net methods of forecasting key finance figures. The author focuses at digital transformation of company oriented to the use of micro-service architecture and cloud technologies but he also underlines that neuron nets have not been applied so far to forecast proceeds in Russian practice. The research is based on analyzing data since 2019 after the merger of the companies ‘M. Video’ and ‘Eldorado’. The use of correlation-regressive analysis for forecasting proceeds of the organization showed a low prognostic accuracy and economic inadequacy of results. In response a series of neuro-net models were developed on the basis of Deductor Studio Lite 5.1 with method of sliding window, including bayes ensemble of five multilayer perceptrons of different architecture. All models demonstrated high accuracy of approximation (mean error is less that 0.01%) and the best results were reached by the two-layer net (6 and 8 neurons in concealed layers). A conclusion was drawn that neuro-net models exceed traditional methods by accuracy and sustainability of forecasting and their introduction into practice of trade company management could provide a promising line in further digital transformation.