Modelo de Scoring para crédito de consumo en la Cooperativa de Ahorro y Crédito “Minga” Ltda utilizando técnicas de Machine Learning

Abstract

The objective of the research has been to create a scoring model for the savings and credit cooperative "MINGA" Ltd, which allows the efficient choice of the consumer credit client. For this study, the information has been obtained directly from the cooperative and the analysis has been done with a database of 15108 records. Subsequently, a model was established and good and bad customers were defined based on days past due. For this, the data has been prepared, so the data set has been reduced to 9065 records and 14 variables were considered. Once the information was ready, it has been applied in the modeling of Machine Learning techniques by using the decision tree method, with the Random Forest algorithm based on bibliographic recommendations, training it with 100, 500 and 1000 trees, with and without selection of variables, with and without categorization of qualitative variables. The results aimed to demonstrate the comparison of the precision of the models that were applied in the study, evidencing as the winner model 2 with 13 variables, with selection of characteristics, without categorization of the qualitative variables with 500 trees and an accuracy of 97.85%, having an error rate of 2.15%. Concluding that a good payer has a score greater than or equal to 600, income of 1000, and expenditure of 250 dollars, between 25 and 40 years of age, it is recommended including new categories in the class.