Customers churn prediction with rfm model and building a recommendation system using semi-supervised learning in retail sector

Abstract

Customer churn or customer attrition occurs when certain customers are no longer loyal to a firm. In retail businesses, the event of churn is said to occur, if a customer's transactions terminates after a certain duration. High churn rates incur humungous losses for the businesses as it is observed that acquiring new buyers is costlier than retaining the current customer base. Hence, for calculating customer churn of companies, they should be able to monitor churn rates. These churn rates give an organization various factors to be considered to determine their customer retention success rates and identify strategies for improvement. Customer churn is predicted using Pareto/NBD model. Once the customers who are likely to churn are predicted, they need to be differentiated based on their previous purchasing history. Natural Language Processing is used to model product categorization. Semi-supervised learning does customer segmentation. This consists of assigning a score by RFM model and segmenting using k-means clustering. The prediction of clusters is then done using algorithms like logistic regression, SVM and SGD classifier. These methods are collectively used to build a suitable recommendation system, which is targeted to make the churn customers who were valuable to the company loyal again, thereby improving the business for retailers.