FORECASTING OF CREDIT CLASSES WITH THE SELF-ORGANIZING MAPS

Abstract

To determine the credit classes statistical and artificial intelligence methods have been used often recently. Particularly the artificial neural networks there have been often applied, one of them is a self-organizing map (SOM). SOM is a two-dimensional map of the credit units that is generated by similar characteristics (attributes) of the process. However this process is not specified by network outputs. If the credit units of one class dominate in the clusters it is valuable such SOMs to employ to forecast the new credit classes. In this paper we investigate the capabilities of SOM in forecasting of credit classes. We present the results of our investigations and show that SOM may distinctly reduce misclassification errors. On the other hand, we demonstrate the possibility of SOM to identify how dataset is liable for the valuable map generation.