A Development of Predictive Model for Customer Retention using Machine Learning Techniques: A Case Study of a Parcel Service Franchise

Abstract

   This research aims to 1) develop a predictive model using machine learning techniques to forecast customer retention, 2) compare the performance of various machine learning models, and 3) analyze key determinants influencing customer re-usage behavior in a parcel delivery franchise. The research methodology was conducted based on the Cross-Industry Standard Process for Data Mining (CRISP-DM) The study utilized empirical transaction data from August 2025 to January 2026. Features extraction was based on the Recency, Frequency, and Monetary (RFM) analysis framework, supplemented by the Cash on Delivery (COD) service. The Tabular Generative Adversarial Networks (CTGAN) was employed to synthesize data and mitigate class imbalance, enhancing model performance. Subsequently, four machine learning algorithms as Logistic Regression, Random Forest, XGBoost, and Support Vector Machine (SVM) were evaluated. The findings successfully addressed the research objectives as follows: 1) The predictive model was successfully developed, effectively mitigating the learning bias caused by data imbalance. 2) The comparative analysis demonstrated that the XGBoost model achieved the highest predictive performance, yielding an Area Under the ROC Curve (AUC-ROC) of 0.8462, thereby proving to be the most suitable algorithm for predicting customer return. 3) The feature importance analysis indicated that the Recency of service usage was the most influential predictor, followed by the COD service usage status. The results suggest that leveraging datasets from small businesses to construct predictive models can serve as an effective decision-support tool. Furthermore, this enables businesses to formulate proactive customer retention strategies, increase re-usage rates, and establish a competitive advantage in the logistics industry.