Orthogonal least square and optimized support vector machine with polynomial kernel for classifying asphyxiated infant cry

Abstract

This paper describes the classification of infant cry with asphyxia using orthogonal least square based support vector machine with polynomial kernel. Optimization of input feature set and filter bank number of mel frequency cepstrum coefficient were performed to produce accurate results. These input feature sets were classified using support vector machine (SVM) with polynomial kernel. To enhance the performance of the classifier, the optimal feature set was then ranked in accordance to its error reduction ratio using orthogonal least square (OLS) and the classification was then repeated. In the experiments, the optimal regularization parameter and polynomial order of 2 were used. It was found that the optimal input feature set for SVM with polynomial kernel is 10 coefficients and 22 filter banks. The highest classification accuracy obtained is 96.06% when OLS is combined with SVM.