Detection of asphyxia from infant cry using support vector machine and multilayer perceptron integrated with Orthogonal Least Square

This paper describes the classification of infant cry with asphyxia using integration of Orthogonal Least Square and Support Vector Machine with Radial Basis Function kernel (OLS-SVM) and integration of Orthogonal Least Square with Multilayer Perceptron (OLS-MLP). The information embedded in the cry signal was extracted using Mel Frequency Cepstrum Coefficient (MFCC) analysis. The extracted features were then selected according to its error reduction ratio (ERR) using OLS. MLP and SVM were then used to distinguish between asphyxiated infant cry and normal cry. Classification accuracy was computed to evaluate the performance of both methods. The OLS-SVM has produced high classification accuracy (94.34%) compared to OLS-MLP when C and γ were set to 1 and 0.013 respectively, and the selection of coefficients is 30% of 33 filter banks.