Classification of respiratory sounds using crackle parameters

The three-class recognition problem of respiratory sounds based on spectral estimation is addressed. Respiratory sounds of two types of pathological cases, namely, obstructive and restrictive disease patients, and healthy subjects are used to obtain feature parameters by dividing each respiratory cycle into overlapping segments and applying an ARMA model. Furthermore, crackle parameters are added to the feature space to observe whether an improvement is achieved in the classification. In this work, k-NN and multinomial classifiers are used in accordance with previous work