Predicción de Churn Voluntario en plataforma de Comercio online
Loading...
Date
Authors
relationships.isAdvisorOf
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad Torcuato Di Tella
Abstract
Este trabajo aborda la problemática de retención de clientes en el contexto de una empresa
que ofrece servicios digitales a otras empresas comerciales.
Dado que el servicio funciona en un esquema de comisiones y no de suscripción, se plantea
un enfoque de prevención de churn transaccional, en el cual el foco es que el cliente
continúe operando con la empresa.
Para ello, se identifican y analizan las principales variables relacionadas a los clientes que
abandonan la plataforma. A partir de este análisis, se realiza el entrenamiento de tres tipos
de modelos de aprendizaje automático (Regresión Logística, Gradient Boosting y Random
Forest) y se compara su performance a través de la curva ROC.
Finalmente, se optimizan los hiperparámetros del modelo seleccionado a través de Grid
Search, y se utilizan las predicciones para generar campañas de retención enfocadas en los
grupos de clientes que poseen mayor riesgo de abandono.
This research tries to address the problem of customer retention in the context of a company that offers digital services to other commercial companies. Since the platform works on a commission scheme and not a subscription, a transactional churn prevention approach is proposed, where the focus is on the client continuing to operate with the company. The main variables that are related to customers who leave the platform are identified. From this analysis, three types of machine learning models are trained (Logistic Regression, Gradient Boosting, and Random Forest) and their performance is compared through the ROC curve. Finally, the hyperparameters of the selected model are optimized through Grid Search, and the predictions are used to generate retention campaigns focused on the customer groups that have the highest risk of churn
This research tries to address the problem of customer retention in the context of a company that offers digital services to other commercial companies. Since the platform works on a commission scheme and not a subscription, a transactional churn prevention approach is proposed, where the focus is on the client continuing to operate with the company. The main variables that are related to customers who leave the platform are identified. From this analysis, three types of machine learning models are trained (Logistic Regression, Gradient Boosting, and Random Forest) and their performance is compared through the ROC curve. Finally, the hyperparameters of the selected model are optimized through Grid Search, and the predictions are used to generate retention campaigns focused on the customer groups that have the highest risk of churn
Description
Keywords
Satisfaccion del cliente, Estrategia comercial, Comercio electrónico, Predicción tecnológica