The Credit Approval data set contains 690 instances concerning credit card applications [quin96]. There are 15 attributes describing the applicant; some of the input values are continuous, others are symbolic. The two classes `+' or `-' indicate whether the application was successful. The class distribution set is fairly evenly balanced (44.5% `+' and 55.5% `-').
For the experiment all input variables were converted into numeric values and normalized in the interval .
In the test the performance of a single BP trained MLP was compared to two different modular networks, one with three sub-nets in the input layer and one with two. The best results achieved with both modular configurations was between 84% and 85%. The best result achieved by a single network was 86%.