Cough Event Prediction based on Spectral Features and SVM and KNN Machine learning using Triaxial Accelerometer data from Multiple body positions

Cough is a major illness and needs to be tackled efficiently to manage respiratory distress and the long-term effects of respiratory problems. In this work an efficient method is proposed using the multi-band spectral features of triaxial accelerometer data which is recorded from different body positions is analyzed and this method reduces the computational complexity significantly. The measurements yield better results in the Y axis compared to the X and Z axes for chest and stomach positions whereas for the ear position it is observed that Z axis produced a higher feature score for cough activity prediction. The classification was performed using SVM and KNN and the best performance observed was in the ear position with a maximum accuracy of 99<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>. Unlike prior works that use computationally intensive methods such as CNN the proposed method uses Spectral features and SVM and KNN from worn Triaxial Accelerometer signals.