A credit scoring analysis using data mining algorithms

Credit scoring is very important nowdays as it helps lenders to evaluate new credit applicants, it is an analysis through which banks can decide beforehand if a customer will be able to repay his debt, among with the interest, based on the historic data of former and present debtors. The purpose of this paper is to conduct a comparative study on the accuracy of classification models, the data base is formed of 150 individuals from a local bank in Romania, and there are used 18 dependent variables. The data mining algorithms selected for this comparison of credit scoring models are the following decision trees: CART, CHAID, Boosted Tree and Random Forest. The reason why the above mentioned models were selected in this study and not others is because in recent years decision trees were increasingly used to build credit scoring models, also their results can be easily interpreted and they can be applied on both categorical and continuous data. The results obtained in Statistica software indicate that Random Forest has higher accuracy rates and therefore outperformes the other proposed classification methods when it comes to distinguishing between good payers and bad payers.