A soft computing tool for species classification and prediction of glucomannan content in genus.

The proposed work aims at designing a classification system for automatic identification of species, grown as a potential cash crop in many Asian countries, from the DNA fingerprints of genus. Four sets of 48 DNA fingerprints belonging to 37 species of the genus, developed with the help of four different primers are considered for the experiment, with an objective to identify only the fingerprints of the species of interest. A second experimental setup deals with the automatic classification of species containing high amounts of glucomannan from the same set of DNA fingerprints of the genus. For each set of 48 DNA fingerprints generated with a specific primer, the DNA fingerprints are preprocessed to extract a 42 dimensional feature vector which is used to generate a k-Nearest Neighbor based classifier based on the Leave One Out Cross Validation protocol. Final classification based on outputs from individual classifiers constructed with respect to the four different primers is performed according to a n-star consensus strategy. The n-star consensus predicts species with cent per cent accuracy while it predicts species containing glucomannan with a more modest accuracy of 81.25%.