Machine learning for credit risk in the Reactive Peru Program: A comparison of the Lasso and Ridge regression models
Soria Quijaite, Juan Jesús
Geraldo-Campos, Luis Alberto
MetadatosMostrar el registro completo del ítem
COVID-19 has caused an economic crisis in the business world, leaving limitations in the continuity of the payment chain, with companies resorting to credit access. This study aimed to determine the optimal machine learning predictive model for the credit risk of companies under the Reactiva Peru Program because of COVID-19. A multivariate regression analysis was applied with four regressor variables (economic sector, granting entity, amount covered, and department) and one predictor (risk level), with a population of 501,298 companies benefiting from the program, under the CRISP-DM methodology oriented especially for data mining projects, with artificial intelligence techniques under the machine learning Lasso and Ridge regression models, with econometric algebraic mathematical verification to compare and validate the predictive models using SPSS, Jamovi, R Studio, and MATLAB software. The results revealed a better Lasso regression model (λ60 = 0.00038; RMSE = 0.3573685) that optimally predicted the level of risk compared to the Ridge regression model (λ100 = 0.00910; RMSE = 0.3573812) and the least squares model with algebraic mathematics, which corroborates that the Lasso regression model is the best predictive model to detect the level of credit risk of the Reactiva Peru Program. The best predictive model for detecting the level of corporate credit risk is the Lasso regression model.
Nivel de acceso
- Artículos