Aplicação de regressão logística para previsão de inadimplentes de cartão de crédito

Abstract

Predicting defaults is an important challenge for financial institutions, especially in the credit card industry. In the present work, the application of logistic regression is explored as a technique to predict the probability of default of credit card customers, since this technique is generally used to predict binary events. Using a set of historical data containing clients’ socioeconomic and transactional information, we built a logistic regression model to identify the main factors that contribute to default and evaluate its predictive capacity, via a confusion matrix that compares actual and predicted data in a small table, in addition to calculating some metrics of model fit quality such as accuracy, sensitivity and specificity. Finally, a ROC curve was constructed to evaluate the performance of the classification model as it your threshold increases