Customer lifetime value and churn threshold prediction using machine learning and deep learning techniques

With the acceleration of digitalization after 2020, accurately analyzing customer behavior has become a strategic priority for businesses. In this study, a holistic customer segmentation model tailored to the e-commerce sector is developed by jointly addressing customer lifetime value and customer churn analysis. The model aims to predict both the customers' potential future spending amounts and the spending thresholds that precede churn. A three-stage approach was adopted in the study. Model 1 estimates customers' expenditures for the next six months using regression-based algorithms, based on behavioral and purchase data from the past 12 months. Model 2 investigates the behavioral patterns of churned customers to forecast the spending threshold at which currently active customers are likely to leave the platform. Model 3 clusters customers according to their profitability and loyalty levels by analyzing the predictions obtained in the previous stages using clustering algorithms. Advanced machine learning and deep learning methods such as XGBoost, Gradient Boosting, LightGBM, Random Forest, Artificial Neural Networks, Deep Neural Networks, Recurrent Neural Networks, and Convolutional Neural Networks were tested during the modeling process. For segmentation, K-Means and Hierarchical Clustering algorithms were used. This integrated approach stands out as an alternative to one-dimensional classifications by addressing both CLV and churn behavior within a unified framework. The resulting segmentation reveals meaningful customer groups such as "Loyal and Strategic Customers," "Profitable but High-Risk Customers," "Promising Customers with Growth Potential," and "Low-Risk and Low-Value Customers," offering actionable insights for marketing strategies. Conducted using data obtained from the e-commerce sector, this study provides significant contributions to both academic literature and industry practices through its field-oriented structure and holistic modeling approach.