Evaluation of empirical attributes for credit risk forecasting from numerical data

In this paper, we evaluate 35 features which are empirically utilized for forecasting the credit behavior of the borrowers of a Greek Bank. These features are initially selected according to universally accepted criteria. A data set of historical data (observations) was collected from the database of a Greek bank. Based on those data, we performed extensive data analysis by using non parametric models. Our data analysis revealed that building a simplified model by using only 3 out of the 35 initially selected features can achieve the same or slightly better forecasting accuracy when compared to the forecasting accuracy achieved by a model which uses all the 35 features. Extensive interpretation of the results is provided and the experimentally verified claim that universally accepted criteria can’t be globally used to achieve optimal results is discussed