Respiratory sound classification : different sensing locations

• Main Author: Wee, Ian Thai Yu
• Other Authors: Ser Wee
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/149975
• Institution: Nanyang Technological University

The presence of a respiratory disorder is the cause for some medical problems. Over the years, machine learning algorithms have been developed for the classification of a cardio-respiratory disorders. This paper aims to use samples acquired through sound-based techniques. Through feature extraction of the samples, followed by using classification techniques, the results would be evaluated to determine whether difference in the sensing location affects the classification results. By means of extracting the Mel-Frequency Cepstral Coefficients, the features of sound samples collected from different location of the lung are obtained. The Fisher’s Ratios between the samples of different locations are obtained to determine the most discriminative or useful the features are. The top features are used in the classification method, Support Vector Machines, to classify the samples into their classes. Based on the classification result, the project finds that linear Support Vector Machines is able to classify between class, but it produces a mediocre result. A different kernel might produce better results. This concludes that there is an effect on classification result based on sensing location, however further work is needed to determine whether the effect of different sensing location to classification results is significant.