Selection of variables for credit risk data mining models: Preliminary research

摘要

Credit risk is related to the risk of the borrower that the lender will not be able to return their debt including interest. Numerous researches have been conducted in the area of credit risk, both using classical models such as Altman Z-score and using machine learning methodology. However, the research using the data from Croatian financial institutions is scarce, especially research focused on the selection of the demographic and/or behavior variables. In addition, it is important to develop robust models that estimate credit risk as accurately as possible. The goal of this research is to develop a data mining model for prediction of credit risk, using the data from Croatian financial institutions on defaulted clients (demographic and behavior data). Decision tree models are constructed for the prediction of credit risk. Different algorithms for the variable selection are evaluated based on the classification accuracy of the decision trees developed based on the selected variables.