Two-way analysis of variance (ANOVA) ranking of features based on wavelet bi-phase and bi-spectrum for the classification of adventitious lung sounds

Pulmonary obstruction diseases are contributing to a greater share of deaths of human beings in the world. Patients suffering from these diseases produce adventitious sounds in the breathing cycle. Studies concentrate on features based on signal amplitude, so the detection accuracy depends upon the quality of signals. The pulmonary disorder produces non-linearity in the adventitious sounds. This research targets the non-linear characteristics of adventitious sounds, such as wheezes, crackles, and normal sounds. To test the proposed feature sets, the analysis uses the ICBHI’17 database. The research introduces sixteen features (two carried forward) based on wavelet bi-phase (WBP) and bi-spectrum (WBS). The feature extraction step follows the calculation of probability histograms, and then features are ranked by ANOVA. The research employs decision tree, SVM, k-NN, NN, and ensemble learners with two to five sub-classifiers (Matlab^®2021a, MathWorks^®, Inc.) to analyse the features. Results show that features ranked with ANOVA clubbed with classifier have achieved maximum accuracy of detection as 98.4%.