RF-based respiration disorders sensing and classification using machine algorithms

The advent of real-time wireless sensing technologies holds promise for revolutionising healthcare provision, particularly in incidences requiring continuous monitoring, such as Cardiovascular Diseases (CVD), heart attacks and other infectious diseases affecting the respiratory system. Leveraging Universal Software Radio Peripherals (USRP), this study proposes a Radio Frequency (RF) sensing approach based on experiments to capture respiration data qualitatively. The study methodology involves selecting the frequency subcarrier from USRP raw data, followed by noise removal, data smoothing, and normalisation. Subsequently, relevant features are extracted from the preprocessed data, facilitating the training of Machine Learning (ML) models to enable respiration disorder classification. A comprehensive evaluation of various ML algorithms reveals that Extremely Randomised Trees (ERT) and Multilayer Perceptron (MLP) outperform others in classifying RF-based respiration real-time data, achieving an outstanding accuracy of 100% with comparatively short training duration.