The classification of churn is driven by the potential risks e-commerce companies face, such as losing customers who discontinue their service usage or churn. Marketing specialists have shifted their efforts from acquiring new customers to retaining existing ones in order to mitigate customer churn. Predictive models are created using data mining techniques to identify customer churn patterns. This study proposes a data mining model aimed at predicting customer behavior, with the processed results utilized as suggestions for improvements and company strategies in customer retention through segmentation and classification. Segmentation and classification involve several variables: Session, Interaction with Application, Actions taken during the interaction, purchasing, claim, and discount. This study employs a clustering technique based on the Recency, Frequency, and Monetary (RFM) model, which considers factors such as the time since the last visit, the number of visits, and the total amount spent by the customer. The classification algorithm model was evaluated by comparing three classification algorithms: decision tree and Support Vector Machine (SVM). The decision tree algorithm had the highest accuracy, achieving an impressive 87% accuracy rate in customer classification. Factors influencing customer churn include purchasing behavior, session activity, claim feature utilization, adding products to cart, and discounts. Improving stock management is crucial to prevent stock shortages, likely to cause churn. Additional measures like sending emails/notifications and offering vouchers/loyalty points can be implemented for customers who added products to their carts but didn't complete the purchase, with a focus on popular products.
Read full abstract