Random forest based feature ranking to evaluate the effect of motion artifact on different clinical features of PPG Signal

A finger pulse signal or Photo plethysmogram (PPG) can provide important cardiovascular information about the human body. But one major problem with PPG signal is that it is prone to motion artifact (MA) due to peripheral body movement. Hence pre-processing is an essential prerequisite before its accurate analysis. In this work, we investigate the effect of MA on the clinical features derived from the PPG signal. A moving window-based new peak detection algorithm is proposed in this work. From the detected peak and foot, the clinical features are calculated. A random forest algorithm is used for feature selection, and the features were ranked based on the effect of MA on the feature. The new peak detection algorithm showed more than 40% improvement from earlier works. The random forest feature ranking showed that pulse area and pulse interval are the two features which are mostly affected by the motion artefact whereas systolic amplitude and half systolic peak heights are the least affected features due to motion artefact.