The effect of F-ratio in the classification of asphyxiated infant cries using multilayer perceptron Neural Network

Abstract

Artificial Neural Network has been widely applied for solving pattern recognition problems including infant cry classification for detecting infant health and physical status. Feature extraction is usually performed using Mel Frequency Cepstrum Coefficient (MFCC) analysis. If irrelevant features in the MFCC are not removed, the performance of the MLP will be degraded. The use of F-ratio is essential to select the significant features. This paper examines the effect of selecting features using F-ratio on the classification accuracy of the MLP. Results obtained from direct selection of coefficients and selection of coefficients via F-ratio, were compared. It is found that the contribution of F-ratio in the selection of input for the MLP has managed to produce high classification accuracy.