An empirical study of fuzzy ARTMAP applied to cytogenetics

The fuzzy ARTMAP (FAM) neural network is evaluated in a pattern classification task of discriminating signals identifying genetic diseases. The FAM provides incremental learning necessary to cope with the expansion of genetic applications and variety of biological preparation techniques. Two training modes of the FAM, training until completion and training with validation, are experimentally compared with respect to their accuracy and sensitivity to the vigilance parameter. Although overfilling the training set, the FAM accuracy on the test set after being trained until completion outperforms that achieved utilizing a validation set. This classification accuracy is completed employing less than five epochs compared to hundreds of training epochs required for other neural network paradigms to accomplish similar performance.