Stress Detection with Deep Learning Using BVP and EDA Signals

In daily life, a person is exposed to many negative factors and emotional states as arising from financial difficulties, working life and personal responsibilities. The aim of this study is to detect stress resulted from these factors in a way that causes the least discomfort to the person. It is quite common to use physiological data such as heart rate, electromyography (EMG), electrocardiography (ECG), electroencephalography (EEG), respiration and skin conductivity to detect when a person is exposed to stress. In this study, stress estimation was made using blood volume pulse (BVP) and electrodermal activity (EDA) sensor data obtained from the Empatica E4 device in the open source WESAD dataset. With the use of BVP and EDA sensors, the intervention to the person has been tried to be minimized. Thus, the model ans sensor method proposed in this study can be easily adapted to daily life. For stress detection, a feed forward deep learning artificial neural network (ANN) technique is proposed by using the baseline and stress labeled data in the dataset. With ANN model, %96.26 accuracy was obtained and a fairly smooth loss curve was observed. This model was compared with the ANN methods in previous studies.