The object of research is the process of analyzing the customer churn of telecommunications companies based on machine learning methods. The existing problem is that, until recently, the process of customer churn was compensated by attracting new customers, but in the modern world, growth rates are constantly accelerating, the market is filled with a large number of competitors, which leads to a constant increase in customer requirements for products and services. In this regard, the process of attracting new customers becomes more costly and time-consuming, which in turn enhances the importance of maintaining an existing customer base. The paper considers problematic aspects related to improving the accuracy of predicting the outflow of a company's customers through the use of machine learning methods. The conducted studies are based on the application of an approach implemented by combining the methods of decision trees and nearest neighbors. A positive result cannot be achieved by ignoring the existing resource constraints and requirements, which must be determined separately for each research case. The relevance of the problem of analysis the outflow of customers for companies with many users is considered. A model for predicting the outflow of customers based on a combination of decision tree and nearest neighbor methods, which is used in the basis of the bagging method, is proposed. One of the features of this approach is the use of a test sample of normalized data. Accordingly, systems can use pre-known information, learn, acquire new knowledge, predict time series, perform classification, and in addition, they are quite obvious to the user. The prospect of choosing these methods is explained by the fact that they were used earlier in data analysis systems and provided sufficiently high-quality results. The expediency and prospects of applying the proposed approach in the problem of analysis the outflow of customers of telecommunications companies are shown, as well as the design features of information technology and the results of software implementation.
Read full abstract