A Stress Detection Model Based on LSTM Network Using Solely Raw PPG Signals
Koorosh Motaman, Khalil Alipour, Bahram Tarvirdizadeh, Mohammad Ghamari
Abstract
Everyone experiences stressful situations in daily life. Stress can have effect on mental and physical health. Thus, it is urgent and sometimes vital to diagnose stress instantly to avoid its future destructive effects. To achieve this goal, physiological signals can be used. Physiological signals such as Photoplethysmography (PPG) can be useful for developing models built on Machine Learning and Deep Learning algorithms for stress detection. In this research, exclusively raw PPG signals have been used for developing a LSTM based stress detection model. More specifically, raw PPG signals without extracting additional features have been employed to train the neural network model. The results suggest that, using neural networks, it is possible to detect stressful events. The proposed model in this research was able to diagnose the stressful states with an accuracy of 88.44%. In addition, statistical measures such as AUC (Area Under the Curve) and F1-score have been used to evaluate the model prediction. As a result, the proposed model was able to obtain the AUC of 93.19%, and the F1-score of 88.05%.