Binary Particle Swarm Optimization and F-Ratio for Selection of Features in the Recognition of Asphyxiated Infant Cry

Abstract

AbstractIn the infant cry classification for detecting pathological conditions using Artificial Neural Network, a common feature extraction technique employed is Mel Frequency Cepstrum Coefficient (MFCC) analysis due to its good representation properties. However, not all MFCC features are significant for classification. If irrelevant features are retained, the performance of the classifier will be degraded. This paper examines the performance of F-ratio and BPSO in selecting infant cry features obtained from MFCC analysis. The performance of both methods was evaluated based on the classification accuracy produced when the selected features were passed to Multi-Layer Perceptron (MLP) classifier. It was found that the BPSO managed to produce better result compared to F-Ratio technique. The classification accuracy achieved using F-Ratio was 93.38%, which was obtained when 29 MFCC filter banks, 8 selected MFC coefficients and 45 hidden nodes were used. The BPSO managed to obtain classification accuracy of 96.03% using 34 MFCC filter banks, 16 selected MFC coefficients and 5 hidden nodes of MLP.KeywordsAsphyxiaMel Frequency Cepstrum Coefficient (MFCC)F-ratioMultilayer Perceptron (MLP)Binary Particle Swarm Optimization (BPSO)