Evaluation of Different Machine Learning Models for Photoplethysmogram Signal Artifact Detection

Photoplethysmography (PPG) is a convenient as well as a simple method to detect the change in blood volume level. It is recently in wide use for noninvasive measurement using optical technique. But PPG signals are very sensitive to various artifacts. These artifacts impact measurement accuracy in negative way which can provide a significant number of inaccurate diagnoses. Thus in this paper, we propose to build a system to detect PPG signal artifacts of the MIMIC database and divide them into two classes, one is acceptable and another is anomalous. Different machine learning algorithms were applied to see the classification accuracy. Among them, Random Forest (RF) performed the best with the accuracy +/- standard deviation of 84.00 +/- 2.89%.