Analysis Of Features Based On Wavelet Bi-Spectrum And Power Spectrum For The Detection Of Adventitious Lung Sounds

Lung sounds contribute essential information about a patient's health. Here, for detecting adventitious sounds of lungs, two sets of eight features each (total 16 features), based on wavelet bi-spectrum (WBS) and power spectrum (WPS), respectively, are proposed. The feature sets are analyzed using five classifiers with one to seven sub-classifier types. A matrix (17X14) of seven evaluation parameters compares the feature sets. Results show that Random Forest, Random Tree, Random committee with WPS and WBS features, LMT with WPS, and Randomizable filter classifier with WBS have shown the best results updating the accuracy obtained in previous researches.