Rentabilización de clientes por medio del aprendizaje automático

Abstract

This work focuses on predicting the desertion of a client within a certain time in the banking sector in order to establish a retention strategy. To achieve this objective it is necessary to understand the business context in which the problem develops and the type of defection that will be predicted. The techniques that were used are vector support machines and logistic regression in order to verify the accuracy of machine learning versus a traditional technique. Since the attrition problem has data imbalance, undersampling was used in the data. Once the probability of desertion has been calculated, it is calculated in value of the client's life time to determine if it is worthwhile or not to retain it.