Modelo de predicción de la probabilidad de pago de los clientes de una institución financiera mediante técnicas de Machine Learning

Abstract

Abstract: This thesis presents a predictive model using Machine Learning to estimate customers payment probability in a financial institution. A real dataset of 10,000 client records with 36 socio-financial variables was used. Following the CRISP-DM methodology, the project conducted business understanding, data preprocessing, and feature engineering, then trained supervised classifiers e.g. logistic regression, SVM, decision tree to predict default risk. The models reveal key payment behavior patterns and classify customers by likelihood to repay. The resulting tool enables the bancking institution to target high-risk segments with focused debt-collection campaigns, improving decision-making and resource allocation for portfolio recovery.