| Resumen |
Nowadays, credit assignment constitutes a way in which persons or entities access to money. However, bad clients can cause big distress to financial institutions. If there are appropriate data banks whose patterns contain financial information from the scope of the allocation of credits, the intelligent pattern classifiers are ideal candidates to solve the credit assignment problem. Nevertheless, working with data sets from credit environment has the disadvantage that, in most of the cases, have unbalanced classes. This situation represents a problem at the moment of work with this kind of datasets due to the fact that unbalanced classes, in general, create biased learning. The consequences of this are reflected during the testing phase because the biased learning causes the classifiers to just recognize appropriately the elements of the ruling class and therefore, give us inaccuracy results. In this paper, we tested some undersampling and oversampling algorithms, and we compared their performance, based on the Imbalance Ratio measure, over different well-known credit related datasets. |
