Stress Level Prediction Using Data From Wearables

Nowadays, stress is one of the main concerns when it comes to people's well-being and health. Knowing beforehand which events may cause stress, can be helpful in managing it. This paper aims to provide a solution for predicting stress levels caused by future events. We are proposing a system targeted at a representative test scenario for this problem, which uses calendar events and heart-rate variation data from wearables to offer predictions on stress levels for future events. This test scenario has a wide applicability as the majority of individuals in modern societies have desk jobs, and a significant part of their daily activities is recorded in their digital calendars. The system consists of a microservices data gathering infrastructure and a random forest algorithm for correlating heart-rate variations to calendar events. A dataset including data from 2 employees of a tech company over a period of one year, having attended a total of 1053 meetings, was used for evaluating the proposed system. The accuracy of the heart-rate sensors used in this study was validated against a professional electrocardiogram device. Our selected machine learning model for predicting heart rate variations for future events, trained on only 400 events, shows errors of less than 12 heart beats per minute 90% of the times. This represents a good performance considering the high heart rate variation when exposed to stress. Furthermore, we have performed our own validation experiment for the correlation of heart rate to stress, which revealed a strong positive correlation of 0.793. This offers us further confirmation that our system is capable of providing indicative estimations of stress levels for future events.