Leveraging gradient boosting machine learning models to predict customer churn in digital health platforms

Customer churn is a critical challenge for sustaining revenue and service continuity in the digital healthcare sector. Accurate churn prediction enables healthcare organizations to implement proactive retention strategies, reduce costs and enhance patient engagement. This study compared the predictive performances of three advanced gradient boosting algorithms, XGBoost, LightGBM, and CatBoost, using real-world behavioral, demographic, and transactional data from a digital health platform. Data preprocessing included handling missing values, categorical encoding, and class imbalance adjustment, followed by hyperparameter optimization using the Optuna framework. The results showed that XGBoost slightly outperformed the other two models, whereas optimization significantly improved the overall performance and stability of all algorithms. Feature importance and SHAP analyses revealed that the average session duration, engagement frequency, and transactional behavior were key predictors of churn. The findings confirm that ensemble gradient boosting techniques offer robust, interpretable, and practical predictive tools for reducing churn and enhancing retention in digital healthcare systems. This study contributes to sustainable health service management by supporting data-driven decisions that promote user retention and deliver high-quality care.