Customer Lifetime Value Prediction with K-means Clustering and XGBoost

Abstract

Customer lifetime value (CLV) is the revenue expected from a customer over a given time period. CLV customer segmentation is used in marketing, resource management and business strategy. Practically, it is customer segmentation rather than revenue, and a specific timeframe rather than entire lifetimes, that is of interest. A long-standing method of CLV segmentation involves using a variant of the RFM model - an approach based on Recency, Frequency and Monetary value of past purchases. RFM is popular due to its simplicity and understandability, but it is not without its pitfalls. In this work, XGBoost and K-means clustering were used to address problems with the RFM approach: determining relative weightings of the three variables, choice of CLV segmentation method, and ability to predict future CLV segments based on current data. The system was able to predict CLV, loyalty and marketability segments with 77-78% accuracy for the immediate future, and 74-75% accuracy for the longer term. Experimentation also showed that using RFM alone is sufficient, as augmenting the features with additional purchase data did not improve results.