Graph Signal Processing Based Classification of Noisy and Clean PPG Signals Using Machine Learning Classifiers for Intelligent Health Monitor

Photoplethysmography (PPG) signals play an important role for automatic measurement of pulse rate, blood pressure, non-invasive blood glucose level and respiration rate. Most of the PPG monitoring devices are prone to motion artifacts and noises under different PPG recording conditions. Thus, automatic assessment of PPG signal quality is most essential for discarding unacceptable PPG signals and reducing false alarms due to the noisy measurements. This paper presents a new PPG signal quality assessment (SQA) method by using the average degree feature extracted from the horizontal visibility graph (HVG) of the PPG signal and six different classifiers such as random forest (RF), Naive Bayes (NB), decision tree (DT), support vector machine (SVM), multilayer perceptron (MLP), convolutional neural network (CNN). On a wide variety of standard databases, evaluation results show that the CNN based SQA method had an overall accuracy of 99.24% that outperforms other five SQA methods in terms of overall accuracy. The NB based SQA method had an accuracy of 99.21% with lower memory space of 1 kB as compared to other SQA methods.