Modelos estadísticos para la estimación de la capacidad de pago de personas naturales tarjetahabientes con información en el sistema de registro de datos crediticios : modelo para la estimación de la capacidad de pago en una población de tarjetahabientes.

Abstract

This curricular integration project aims to develop parametric and nonparametric statistical models for the accurate estimation of repayment capacity among banked individuals using information from the credit registry system. Income estimation methodologies were executed using the following models: Multiple Linear Regression (MLR), Random Forest (RF), Gradient Boosting Machine (GBM), and Extreme Gradient Boosting (xg-Boost). In the present project, the database consists of a random and representative sample. To study the database, it was divided into three groups to enhance prediction quality, considering a model for each group. To select the most representative variables for inclusion in these models, Kolmogorov-Smirnov (KS) techniques were applied for quantitative variables, and the Value of Information (VI) technique was used for qualitative variables. Additionally, the resampling technique was a potent tool for improving model predictions, compared to the initial models. The outcomes of this study were favorable according to the objectives. Income estimation was enhanced using the resampling technique compared to the initial model, particularly improving estimation in the tails of real income distributions (individuals with very low and very high incomes).