Analysis of fuzzy logic methods for forecasting customer churn

Abstract

The object of research is the process of predicting the churn of customers of telecommunications companies based on fuzzy logic and neural networks. The research carried out is based on the application of an approach that is implemented through the combined use of fuzzy logic and neural networks. The main assumption of the study is the hypothesis that the use of a fuzzy neural network formed on the basis of fuzzy logic algorithms can improve the accuracy of predicting customer churn relative to available solutions. This result can’t be achieved neglecting the existing resource constraints and requirements, which must be determined separately for each case of research. The relevance of the problem of forecasting customer churn for companies with a large number of users is considered. A model for predicting customer churn is proposed based on the combined use of fuzzy logic and neural networks. The main feature of this approach is that a test sample of normalized data is used at the basis of fuzzy neural networks, which are processed to form the parameters of membership functions that correspond to the inference system, that is, conclusions are made on the basis of a fuzzy logic apparatus. Also, to find the parameters of the membership function, neural network algorithms are used. Such systems can use previously known information, learn, gain new knowledge, predict time series, perform image classification, and besides, they are quite visual to the user. The application of methods of fuzzy logic is considered, they make it possible to obtain a result in the form of a fuzzy inference. The expediency of choosing these methods is explained by the fact that they were previously used in fuzzy automatic control systems and showed sufficiently high quality results. The expediency and prospects of using the proposed approach in the problem of predicting the outflow of customers of telecommunications companies are shown, and the results of software implementation are presented.