oa A Machine Learning Approach for Detecting Mental Stress Based on Biomedical Signal Processing

Mental stress occurs when a person perceives abnormal demands or pressures that influence the sense of well-being. These high demands sometimes exceed human capabilities to cope with. Stressors such as workload, inflexible working hours, financial problem, or handling more than one task can cause work-related stress which in turn leads to less productive employees. Lost productivity costs global economy approximately US $ 1 trillion per year [1]. A survey conducted among 7000 workers in U.S. found that 42% had left their job to escape the stressful work environment [2]. Some people can handle stress better than others, therefore the stress symptoms can vary. Stress symptoms can affect the human body and make him down both physically and mentally. Hopelessness, anxiety, and depression are examples of emotional symptoms, while headaches, over-eating, sweaty hands, and dryness of mouth are physical signs of the stress. There are also behavioral cues for stress like aggression, social withdrawal, and loss of concentration [3]. When the thread is perceived, a survival mechanism called «fight or flight response» will be activated to help the human body to adapt the situation quickly. In this mechanism, the central nervous system (CNS) asks adrenal glands to release cortisol and adrenaline hormones, which boost glucose levels in the bloodstream, quicken the heartbeat, and raise blood pressure. If CNS does not succeed to return to normal state, the body reaction will continue which in turn increases the possibility of having heart stroke or attack [4]. There are several techniques used to explore physiological and physical stress measures, for example, electrocardiogram (ECG) measures the heart»s electrical activity, electroencephalography (EEG) records the brain»s electrical activity, electrodermal activity (EDA) or galvanic skin response (GSR) measures the continuous variations in the skin»s electrical characteristics, electromyography (EMG) records electrical activity in muscles, photoplethysmography (PPG) estimates the skin blood flow, and Infrared (IR) tracks eye activities. On the other hand, prolonged ongoing worrying can lead to chronic stress. This type of stress is most harmful and has been linked to cancer, and cardiovascular disease (CVD) [5]. Therefore, several approaches were proposed in an attempt to identify stress triggers and amount of stress. Some of these methods used instruments such as questionnaires to assess affective states, but these techniques usually suffer from memory and response biases. However, stress detection via the analysis of various bio-signals are deemed more valuable and thus have been the focus of modern day research. In particular, various bio-signals are collected from participates. These bio-signals are then subjected to advanced signal processing algorithms in an attempt to extract salient features for classification by machine learning algorithms. In our project, we are interested in exploring new machine learning techniques which wearable devices to record various bio-signals. The goal is the development of an automatic stress detection system based on the analysis of bio-signals through the use of signal processing and machine learning. The outcome of this research will allow users to be notified when their bodies enter a state of unhealthily stress levels so that they may take preventative action to avoid unnecessary consequences.