Modelo de predição para dados desbalanceados utilizando informações de financiamentos de veículos

Abstract

For the statistics thesis project, a model adjustment is proposed for unbalanced data, using vehicle financing information, with the response variable being dichotomous, divided into defaulters and non-defaulters. Techniques for variable selection, such as weight of evidence and information value, will be presented, along with the adjustment of logistic regression models for both unbalanced and balanced data, model quality metrics, and an interpretable final classification. The project was developed using the Python programming language.