Comparison of Machine Learning Models for Spectral Binning Approach to Motion Classification in Doppler Radar Physiological Monitoring

When using Doppler radar to monitor physiological motion such as that from cardiopulmonary activity, identification of the subject’s state of activity is essential in order to properly assess the measurement. This research examines the use of machine learning (ML) to classify a subject’s state of activity as sedentary, fidgeting, or ambulatory. It adopts a spectral binning approach which classifies components of the arctangent demodulated output signal from a quadrature 2.4-GHz CW Doppler radar system, based on spectral content associated with various frequency bins. Comparative performance was assessed for using Random Forest (RF) and Deep Neural Network (DNN) approaches varying in layers, neurons, and training batch sizes. The most accurate technique proved to be the RF method (93%). However, the other DNN methods also proved effective offering significant potential for robust automated assessment of a monitored subject’s state of activity.