Study of HRV Parameters for Detection of Stress using Machine Learning

Stress is a physiological and psychological reaction to life's circumstances. Stress affects everyone knowing no boundaries. For the short term, stress can be beneficial but ifstress levels are elevated for a prolonged time then it can trouble the body. Studies says that stress influences all age groups between 5 and 49 according to National Centre for Health Statistics 2019. Stress comes in many forms such as Work stress, Depression, Exercise stress, Behavioral stress, Physiological stress, Emotional stress, and diseased condition. 48 % of people have trouble sleeping because of stress and 73% of people's stress impacts their mental health says the American Institute of stress. The existing technology for stress detection involves analysis of ECG, Galvanic skin resistance, glucose level, body temperature, and salivary cortisol which demands patient isolation and medical personal assistance and are influenced by the external environment. Time domain measures of HRV are proved to be potential markers of stress [1]. In this study, we employed two datasets: the Haeley's driver dataset[22] from Physionet and the SWELL dataset from Kaggle. TPOT, an autoML tool that explores millions of pipelines anddetermines which algorithm is the best fit for the data, was used for Healey's driver dataset and a model was developed and the results were evaluated. To find the best model that accurately classifies stress in the SWELL dataset, 9 algorithms were used, models were developed, and cross-validation was performed to find the best model. Finally, all the model results were evaluated and their performances were compared When compared to the other nine models, Random Forest revealed greater accuracy and, upon validation, proved to be a good fit model.