Optimal features for classifying asphyxiated infant cry using support vector machine with RBF kernel

• Published in: IFMBE Proceedings
• Main Author: Sahak R.; Lee Y.K.; Mansor W.; Yassin A.I.M.; Zabidi A.
• Format: Conference paper
• Language: English
• Published: 2011
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-80555148855&doi=10.1007%2f978-3-642-23508-5_27&partnerID=40&md5=1774b657ea76e43bdeef76c2d843e54f

An investigation into optimizing the input feature set for classifier to identify infant cry signals with asphyxia is presented in this paper. Mel frequency cepstrum coefficients were used to represent the infant cry signals collected from the Instituto Nacional De Astrofisica Opticay Electronica, Mexico. Then the number of coefficients and filter banks for the MFC analysis were varied to rank the input feature sets based on the classification accuracy attained and the number of support vector employed. The input features were then used for classification by a support vector machine with radial basis function kernel. A regularization parameter of one and gamma of 0.009, found optimal from our previous experiments, were used. From the results, it was found that the optimal input feature set was obtained with 10 MFC coefficients and 22 filter banks, with a classification accuracy of 93.84% (highest) and a support vector number of 353.1, close to the least of 350.1. © 2011 Springer-Verlag Berlin Heidelberg.