Presents the use of an established data mining methodology as a basis to propose a methodology adapted to credit scoring. Data mining concepts are translated and mapped to the problem and a real case study of a Mexican financial institution is solved using logistic regression, a decision tree, and a neural network. Proposes to use the Brier Score to evaluate the results of the models with the current population. Results show the neural network as the best technique in most metrics using the validation population. However, the Brier Scores show that the logistic regression is more stable to changes in the characteristics of the population.
Mejia, Marcelo; Cadena, Francisco J.; Carrera, Ernesto; and Heredia, Victoria H.
"Desarrollo y Evaluación de Modelos de Calificación Crediticia en una Institución Bancaria Mexicana,"
Revista Latinoamericana Y Del Caribe De La Associacion De Sistemas De Informacion:
1, Article 3.
Available at: http://aisel.aisnet.org/relcasi/vol4/iss1/3