Entropy measurement as features extraction in automatic lung sound classification

Lung sound is one of the important information in the diagnosis of respiratory disease. Many researchers have developed various algorithms to diagnose lung disease through the lung sounds. One of the parameters used as the feature of lung sound is entropy, a measure of the signal complexity in which the normal biological signal and the pathological biological signal have different complexities. Entropy measurement has some different methods performed in a different signal domain as well. This paper discusses the use of various entropy measurements for a lung sound classification. The evaluation results have demonstrated that the use of a single entropy as a feature on the lung sound classification could not produce the high accuracy. The incorporation of 7 entropies, in contrast, could produce the higher one. For five classes of lung sound data, the combination of 7 entropies as features generated the accuracy of 94.95% using multilayer perceptron as a classifier.