Stress Detection and Control According to the Skin Signal of Electrical Resistance and Heart Rate Using Reinforcement Learning

The anxiety disorders are the most common mental health condition. Through modulation of stress responses, the anxiety can be controlled. Stress clearly affects the autonomic nervous system and leads to excessive sweating and tachycardia. New methods to identify and manage stress are still needed.
In this paper, anxiety is controlled by reinforcement learning (RL). For stress level measurement, galvanic skin resistance (GSR) and heart rate were measured. Skin electrical resistance and heart rates are measured in 15 subjects (8 females and 7 males), ages between 22 to 30 years, during exposure to melodies and pictures. By connecting bioelectric sensors to the hand, GSR signal can be tested. Anxiety level was measured by biological signals (biofeedback) through playing melodies and showing pictures. First, GSR and heart rate signals were measured (states of RL). Then, the application plays a melody (agent) and shows a picture (actions). After a period of time, GSR and heart rate were measured again. Based on the differences between initial and secondary values, the reward, indicating stress variations were calculated. Based on the received reward, the agent learns to control the stress level.
The intelligent agent of RL, learns which picture to show and what melody to play in order to manage the stress. For each person, different melodies and pictures reduce the stress. This learning procedure should be repeated for each subject.
A novel method is proposed and investigated based on RL for anxiety control. The accuracy of diagnosis is improved in a simple, rapid, and low-cost way.