A study for the feature selection to identify Giemsa-stained human chromosomes based on artificial neural network

Many studies in computer-based chromosome analysis have shown that it is possible to classify chromosomes into 24 subgroups. In addition, artificial neural networks (ANNs) have been adopted for the human chromosome classification. It is important to select the optimum features for training the neural network classifier. We selected some features - relative length, normalized density profile (d.p) and centromeric index - used to identify chromosomes and trained the neural network classifier by changing the number of samples which were used to get the d.p. We found the fact that the classification error was shown to be at a minimum when this number was equal to or greater than the length of the no.1 human chromosome.