Resumo: The decision making process of evaluating the creditworthiness of a loan is sometimes difficult to the human mind because of the great number of variables and interrelations among them. What we propose here, is to identify the characteristics related to high and low risk, and this is made by using an applicant model. So, with a credit card database with categorical and continuous variables, in order to make the decision process more streamlined and quantifiable, we performed a binary logistic model. Applying the non-hierarchical clustering method (K-means) to the logit output vector we identified eight risk classes. Each class was evaluated temporarily by the pro-duct-limit estimators (Kaplan-Meier estimators) for 70 months, showing that low probability of default is indeed associated with low risk classes. The statistic technique applied to identify the client risk characteristics was the correspondence analysis.
