Maestría en Analítica Aplicada · 2026

Predicción de churn en medios de pago: un enfoque basado en machine learning para la retención de clientes

Resumen

This study addresses the problem of customer churn in an electronic payment platform, where the loss of active merchants directly affects business sustainability. The analysis was conducted using historical transactional data and merchant characteristics from Kiire, which were cleaned, transformed, and prepared for modeling. Several supervised classification algorithms were evaluated, including logistic regression, support vector machines, k-nearest neighbors, and tree-based models. The modeling process was carried out using two AutoML frameworks, PyCaret and NaiveAutoML, in order to assess the robustness and consistency of the results under a homogeneous evaluation scheme. The results indicate that Random Forest–based models achieved the best predictive performance. The final model reached a weighted F1-score close to 0.77 and an AUC of approximately 0.85 on the test set, demonstrating strong discriminative ability for identifying merchants at risk of churn. These findings provide a solid foundation for the development of data-driven retention strategies and highlight the potential of predictive analytics to support decision-making in the electronic payments industry.