Title: A maximum margin classifier for non-linearly separable pattern classes, using a feature space sampling technique, applied to chromosome classification

Authors: S. Ganesh Vaidyanathan, Bibhas Kar, N. Kumaravel

Addresses: Department of Electronics and Communication Engineering, Sri Venkateswara College of Engineering, Sriperumbudur 602105 (TN), India. ' Department of Medical Genetics, Apollo Hospitals, Greams Lane, Chennai 600006 (TN), India. ' Department of Electronics and Communication Engineering, Anna University, Guindy, Chennai 600025 (TN), India

Abstract: The classification of chromosomes using a classifier is generally inaccurate owing to closeness of features belonging to various chromosomes which poses a linearly inseparable problem. This paper proposes a novel technique to obtain the non-linear decision boundary. An average classification accuracy of 93% was achieved with this technique which involves arriving at the non-linear decision boundary by joining and smoothening the sample points obtained by sampling the feature space within a boundary limited by the range of the data and by the curves of the best fit to the two classes. The technique works for feature space of any dimension.

Keywords: feature space; curve fitting; sampling; optimal boundary points; nonlinear boundaries; binary pattern classifiers; chromosomes classification.

DOI: 10.1504/IJBET.2010.034518

International Journal of Biomedical Engineering and Technology, 2010 Vol.4 No.2, pp.123 - 133

Published online: 07 Aug 2010 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article