A multilayer perceptron (MLP) neural network (NN) was used in this study for human chromosome classification. A feature selection technique was used to evaluate the significance of the considered features to the classification results. The technique we used emphasized the status of the centrometric index and of the chromosome length, as the most significant features in chromosome classification. It also yielded the benefit of using only about 70% of the available features to get classification

more »

... erformance close to the ultimate one. The MLP classifier learning curves were examined by measuring the probability of correct test set classification for an increasing size of training sets. Only 10-20 examples were required for the MLP NN classifier to reach its supreme performance disregarding the number of features used. It was also found that the empirical dependence of the entropic error of the classifier on the number of examples is highly comparable to the 1/t function that is a universal learning curve.