Segmentation Analysis and Customer Churn Prediction in Delivery Services Using RFM, K-Means, and Random Forest for Retention Strategy Planning

Abstract

Fierce competition in the delivery and logistics service industry demands targeted customer retention strategies to prevent customer churn that threatens revenue margins. However, customer management practices typically treat customer value segmentation and churn prediction as separate, isolated tasks, leading to inefficient allocation of retention resources. This study aims to integrate transaction data-driven descriptive and predictive modeling to identify customer value segments, predict churn risk, and formulate a prescriptive retention strategy matrix at the JNE Binjai Timur 2 Agency. The secondary dataset comprises 19,389 clean transactions from 4,625 unique customers spanning January 2, 2024, to December 31, 2025. We extracted, log-transformed, and standardized Recency, Frequency, and Monetary (RFM) features using StandardScaler, then applied unsupervised K-Means clustering. To address churn imbalance (defined as inactivity >90 days), class weighting was applied during Random Forest Classifier training. The K-Means clustering results (k=2, Silhouette Score = 0.512) validated two value tiers: Cluster Tier 1 (High-Value, 17.47% of customers) accounted for 69.47% of total revenue (IDR 444.08 million), while Cluster Tier 2 (Low-Value, 82.53% of customers) contributed 30.53% of revenue. The Random Forest model showed superior discriminative performance on the test set (N = 822), with an AUC-ROC of 0.830, Accuracy of 91.61%, Recall of 99.59%, and F1-Score of 95.52%, with Active_Days (0.1580) and Recency (0.1312) as the dominant predictors. Integrating value tiers and churn predictions revealed potential financial exposure within the High-Risk group of IDR 243.52 million (38.09% of total revenue). This study yields four managerial retention intervention priorities (P1–P4), including immediate win-back offers for High-Value High-Risk customers and priority VIP/pick-up services for High-Value Low-Risk customers.