Neural network technologies in supply chain management: Consumer selection technique

The supply chain management’s effectiveness depends, among other things, on the selection and coordinated interaction with product consumers. This article is devoted to the development of a method for selecting a consumer in the regional wholesale and retail fuel market. The methodological basis of the study is the theory of statistical analysis and neural networks. The main tool for developing the methodology was neural network technologies, with the help of which it is most likely possible to correctly estimate the boundaries for indicators’ values that characterize consumers and reflect their history of purchasing behavior, to select potential clients and the possibility of further cooperation with existing ones. The information base for the work is the data on consumers of a given company’s products, data from the 2GIS electronic directory, as well as the results of the primary statistical analysis and forecasts made based on neural networks of various topologies. The author presents his methodology for selecting a consumer. It has the potential for development and implementation for solving a number of other management problems. As part of the testing, the best configuration (topology) of the neural network was determined, and standard values of entry barriers when consumer choice accomplished were assessed. The methodology we developed was tested using the example of a company operating in the wholesale and retail fuel market in Novosibirsk and the Novosibirsk region. When verifying the neural network model, the quality of client classification was compared based on logistic regression, decision tree and random forest models and we found that the neural network approach provides the best results for assessing the degree of client suitability. As a result of testing the methodology, recommendations for improving neural network models were developed, including expanding the set of factors that determine the characteristics of consumers, as well as optimizing the internal structure of neural networks.