E-Commerce Customer Churn Prediction Leveraging Machine Learning and User Behavior Insight

Given the rapid growth and consequently, the fierce competition in the realm of E-commerce, customer churn or attrition has become of vital value as gaining one new customer can cost much higher comparing to retaining the existing customers. Therefore, the more accurate the businesses can predict the churn, the more effective measures they can take on towards customer retention. That is why, this paper presents an innovative approach for customer churn prediction by means of machine learning techniques, specifically Random Forest, Decision Tree, XGBoost, Logistic Regression as well as Naïve Bayes with the perspective of RFM (Recency, Frequency, Monetary) analysis fortified with a prediction enhancement technique, namely Recursive Feature Elimination (RFE). Using a dataset of historical customer data and churn labels, we evaluate the performance of these models based on accuracy, precision, recall, AUC ROC and F۱-score. In addition, it employs RFM to add the user behavior in creating the target variable, namely Churn. Experimental results demonstrate the promising performance of all the models, with XGBoost consistently outperforming others with RFE that had little to no effect. The findings significantly contribute to businesses in implementing proactive churn prevention measures, leading to improved customer retention and profitability.

Keywords:

Machine Learning , Customer churn prediction , RFM , Feature Selection