3-way respiratory sound classification using machine learning algorithms

Abstract

With the development in the fields of artificial intelligence and machine learning, such as Support Vector Machine (SVM) and Convolutional Neural Network (CNN), many biomedical and healthcare applications have gained additional support from high-tech. Nowadays, more and more fields about health problem detection are using sound information and Artificial Intelligence (AI) techniques rather than only depending on image detection or diagnosis from doctors, which could save medical resources as well as improve diagnostic rate and diagnosis accuracy. This project mainly focuses on two types of classification algorithms, one is to project the processed audio signal to a set of spatial points and use geometrical distance to divide them into three parts, the other is to use machine learning algorithms. For geometrical distance, I tried three types of distances: Minkowski, Manhattan and Euclidean distance; for machine learning algorithms, I tried two widely used methods: Support Vector Machine and Convolutional Neural Network, while the SVM part I tried both Gaussian Core and linear SVM. In conclusion, the linear SVM provides the best performance. This dissertation also shows the signal processing progress before the respiratory sound could be classified, which includes audio value extraction, data choosing, signal processing, segmentation, feature extraction, etc.