Stress Detection and Classification from PPG Signals Recorded in Different Conditions - Pilot Study

The paper describes an experiment using a two-level classifier for stress detection and classification from photoplethysmography (PPG) signals acquired in different sensing conditions. In the first phase, the basic decision between the normal state and the joined stress group is performed. The second phase realizes classification among the final stress types. Detection/classification is based on 2D maps of mutual positions of PPG wave features which are in concordant or discordant relationship. The auxiliary experiments demonstrated heavy dependence of the classification accuracy on properly chosen pairs of PPG wave features. The main experiments confirm that the system works properly with only a small PPG signal data collection used for creation of the group of 2D maps. The obtained overall stress classification accuracy about 94 % for the normal group and 88 % for the stressed PPG signal group are promising in this state of research. However, prior to planned real-time application, some implementation and optimization tasks will have to be solved.