Method of selecting borrowers’ features for credit risk assessment

Granting credit involves certain risks and thus financial institutions try to minimize them using various methods. The aim of this article is to develop a method of selecting borrowers’ characteristics for credit risk assessment. The article uses non-genetic algorithms (such as CFS, ReliefF or Lasso), that perform calculations for all features, with only financial institutions granting the loan selecting the significant features to be assessed. The Holland and NSGA II genetic algorithms also begin with all features and gradually reduce their number in the consequent iterations without compromising the accuracy of feature selection. Only the GAAM genetic algorithm gives the lender the ability to choose the number of features entered into the calculations. During the operation, this algorithm increases the accuracy of feature selection by reducing their number. This significantly shortens the time to analyse loan application while maintaining acceptable accuracy, which results in making the right decision quickly. The features selected by the algorithms are additionally classified by the LDA classifier to demonstrate the correct selection of features by a given algorithm compared to the benchmark, which is the actual percentage of credits granted and then repaid.