A model based inference engine for stress estimation

Abstract

Stress has become a household term for which ascertaining a meaning has become increasingly difficult these days. Physiologically, stress is observed to act through hypothalamus which modulates the autonomic nervous system mainly via sympathetically mediated effects. Utilizing this theory, a model based inference engine was developed for the estimation of stress. A computational model was used to generate a series of synthetic photo-plethysmogram (PPG) signals by varying the model parameters. Now using these artificial generated PPG signals, the inverse problem of estimating the stress parameter ‘FSN’ was solved by a neural network, using Levenberg-Marquardt algorithm. The inference engine was then tested by using real PPG data collected twice (morning and evening) from a set of 13 subjects. As observed in experimental studies, our inference engine was able to replicate the pattern of stress levels i.e., exhibiting high levels of stress in mornings compared to evenings. These results validate the efficiency of the developed inference engine in estimating the stress