Customer Churn Prediction Using the RFM Approach and Extreme Gradient Boosting for Company Strategy Recommendation

Abstract

Customers are vital assets in the growth and sustainability of business  organizations. However, customers may discontinue their engagement with a company and switch to competitors’ products or services for various reasons. This event referred to as customer churn. Losing customers significantly impacts a company's revenue, often resulting in financial decline. Churn events, which are subject to dynamic monthly changes, are further influenced by intense competition and rapid technological advancements. Analyzing customer characteristics is crucial to understanding customer behavior, with metrics such as recency, frequency, monetary (RFM) serving as key indicators of subscription and transaction patterns. The Extreme Gradient Boosting method is applied to address the challenge of classifying churn and non-churn customers. The prescriptive analytics process is carried out to identify the features most influential in prediction outcomes, enabling the formulation of strategic recommendations to mitigate churn problems. The integration of RFM analysis with the XGBoost method provides optimal results, particularly in the third segmentation, achieving an accuracy of = 0.98833, precession = 0.98768, recall = 0.98899, and f1-score = 0.98833. The prescriptive analytics process highlights three critical features, namely city factor, GMV generation, and total customer transaction generation. This findings demonstrate that the segmentation characteristics, data representation, and behavioral approach with RFM analysis have an effect on improving the performance of the model in churn prediction.