Dual-Tree Complex Wavelet Packet Transform and Feature Selection Techniques for Infant Cry Classification

Abstract

A Dual-Tree Complex Wavelet Packet Transform (DT-CWPT) feature extraction has been used in infant cry signal classification to extract the feature. Total of 124 energy features and 124 Shannon entropy features were extracted from each sub-band after five level decomposition by DT-CWPT. Feature selection techniques used to deal with massive information obtained from DT-CWPT extraction. The feature selection techniques reduced the number of features by select and form feature subset for classification phase. ELM classifier with 10-fold cross-validation scheme was used to classify the infant cry signal. Three experiments were conducted with different feature sets for three binary classification problems (Asphyxia versus Normal, Deaf versus Normal, and Hunger versus Pain). The results reported that features selection techniques reduced the number of features and achieved high accuracy.