This task is to predict on a two class problem. The data set contains 304 records, with 13 input variables. The variables were recoded into 28 binary inputs. A comparison was only made between a MLFFN and a modular network because the number of inputs was too small to use a LNN.
The data set was investigated in the project [schm96]. The best result achieved in a long number of tests with the program opti was 88.15%.
Different MNNs using standard parameters (learning parameter , momentum , and the steepness of the activation function ) for learning resulted in a performance of 86%. The training was about four times quicker.