ResearchBib Share Your Research, Maximize Your Social Impacts
Sign for Notice Everyday Sign up >> Login

FORECASTING CUSTOMER FUTURE BEHAVIOR IN RETAIL BUSINESS USING MACHINE LEARNING MODELS

Journal: Scientific Journal of Astana IT University (Vol.10, No. 10)

Publication Date:

Authors : ;

Page : 45-60

Keywords : churn; marketing; customer lifetime value;

Source : Downloadexternal Find it from : Google Scholarexternal

Abstract

The ability to forecast customers' future purchases, lifetime value, and churn are fundamental tasks in business management. These tasks become more complicated when the relationship between customers and business is not contractual. Therefore, the application of an appropriate method of customer analysis influences the efficiency of company cost management in interaction with their customers. The purpose of this paper is to compare existing solutions of customer lifetime value prediction and provide a new way to predict the future behavior of customers with consideration of the drawbacks of previous works. The method should have the following properties: use data that is available in any retail business; take into account that markets are constantly changing; be more precise than existing solutions. In this paper, we proposed the method of identifying customer churn provided a way to analyze customer behavior associated with churn or retention. In order to understand why customers churn, we used eleven customer behavioral metrics. The relationship of used metrics with churn was proved using churn cohort analysis. The results of training of logistic regression and neural network on prepared dataset showed that their forecast accuracy is in the healthy range for highly predictable churn. Based on predicted churn probabilities, we calculated the customer lifetime value in the future period. Our research results on customer behavior in the retail business confirm the hypothesis that customers who make many purchases are less likely to churn than customers who make few purchases. The main uniqueness of this work is the way of finding customer churn, as no such data was provided in the initial dataset. In addition, the minimum amount of data that most retail companies have was used. This enables the proposed methodologies to be applied to a large number of retail companies.

Last modified: 2023-01-20 17:02:03