Applied Machine Learning for Affective Computing – Respiratory Disease Detection Project

Description

Using the dataset found here, we compare performances between a Convolutional Neural Network and a Random Forest Classifier to detect the following respiratory diseases from audio of patients' breathing:

• Asthma
• COPD (Chronic Obstructive Pulmonary Disorder)
• LRTI (Lower Respiratory Tract Infection)
• URTI (Upper Respiratory Tract Infection)
• Bronchiectasis
• Bronchiolitis
• Pneumonia

Notes

The script changebit.sh included in this repository is meant to change the bit depth of the .wav files found in the dataset (for use with scipy's spectrogram module). The files from the dataset originally came as 24-bit files, which are not supported by spectrogram. The script overwrites the files at the existing directory.

Features

We convert the .wav files found in the dataset to spectrograms.

Random Forest Classifier Performance

Convolutional Neural Network Performance

Usage