Churn Prediction Using Machine Learning based on E-commerce Data

The main objective of this study is to accurately predict customer churn in order to improve customer retention strategies and enhance long-term profitability for businesses.

The analysis was conducted by combining the RFM (Recency, Frequency, Monetary) model with various machine learning techniques, based on purchase data from 3,900 customers of an online retail store. The results showed that the Random Forest model demonstrated the best predictive performance with a recall rate of 91.11%. Purchase amount, age, customer location, review ratings, and purchased items were identified as key variables for predicting customer churn. Notably, purchase amount emerged as the most important predictive variable, suggesting a close relationship between customer spending levels and the likelihood of churn. To address the issue of data imbalance, undersampling and oversampling (SMOTE) techniques were applied to improve model performance, thereby increasing the prediction accuracy for the minority class of churning customers.

Customers were segmented into loyal customers (38.5%), medium-value customers (35.7%), and high-risk churn customers (25.8%) based on their RFM scores. This segmentation can help in providing special promotions or personalized services to high-risk customer groups to reduce churn rates.

Comparing various machine learning models, Random Forest outperformed in all metrics including accuracy, precision, recall, and F1 score, indicating its ability to capture complex customer behavior patterns effectively. Moreover, ROC curve analysis showed an AUC value of 0.99, confirming the model’s high accuracy in distinguishing between churning and non-churning customers.

These research findings can be valuable for online retailers in predicting customer churn and developing effective customer retention strategies. Particularly, the high-recall model allows for early identification of customers with high churn probability, enabling the development of tailored marketing strategies. This can lead to improved customer retention rates and ultimately contribute to increased business profitability.

Future research should focus on applying a wider range of variables and cutting-edge algorithms to improve model accuracy, as well as analyzing real-world business case studies to validate the model’s effectiveness. Through these efforts, it is expected that the data-driven decision-making capabilities in the online retail industry can be further enhanced, contributing to the development of customer-centric business strategies.