Stress Detection Using Smartphone Extracted Photoplethysmography

Stress is present in our daily live and can affect anyone. It can be induced by many factors and can lead to many health complications such as depression, anxiety and even heart disease. Multiple research found that stress is highly correlated with physiological signals especially pulse rate variability (PRV). In this paper, we generated multiple machine learning models which uses PRV features extracted via photoplethysmography (PPG). Since multiple studies proved PPG can be performed by smartphones, we further compared the stress detection performance for wearables and smartphone extracted data. Moreover, principal component analysis (PCA) and three different time length were observed to check how it will affect the performance of different models. Results from 30 participant shows that Random Forest algorithm achieved the highest classification accuracy for both smartphone and wearable device. An accuracy of 77.65% and 74.51% was achieved for the wearable and smartphone respectively.