ANALYSIS AND SIMULATION OF CUSTOMER BEHAVIOR WITH THE IMPLEMENTATION OF BETA-WEIBUL DISTRIBUTIONS

Abstract

The purpose of study is to summarize approaches to managerial tasks to prevent customer churn according to time factor. Also we outline wide range of machine learning techniques to predict customer churn and apply different methods to create and test a model to project retention rate. Key issue of every business with customers who periodically pay for subscription is retention rate. It is important to differentiate between different types of customers from the point of view of churning, so manager can apply effective strategy to prevent it. And it is not only about the constant traditional marketing process of customer retention, but also about a comprehensive management policy based on the wide range of new data sources and new methodologies, such as machine learning. According to this approach, retention campaigns nowadays must include: initial investment in analytics capabilities, accuracy of targeting, estimation of action effectiveness and cost. Used methods include analysis, synthesis, data transformation, curve-fitting regression analysis, beta-geometric distribution and beta-discrete Weibul distribution. The results of this study suggest that predicting customer retention with linear regression and training a regression model allows to obtain good results even with a small amount of historical data. However, due to the lack of a unified approach to data transformation in such models for different groups of customers, we justified the feasibility of using probabilistic method for predicting the customer retention rate using the beta-geometric distribution and beta-discrete Weibul distribution with the help of package foretell build for R. We also highlight the importance of splitting data for training and for testing models. The study concludes that even with sufficient historical data on customer behavior, we are not able to create a model that can predict the outflow of customers in all possible cases. Apart from data transformations and methods applied we should consider demographic characteristics of customer group, duration of product usage and other factors. Therefore, it is necessary to test the model and choose your way of training the model. Keywords: Customer behavior, estimation, regression analysis, retention rate, beta geometric distribution, beta discrete Weibull distribution