Relax: Learning Agent for individualised Stress Management

Many studies show a permanently increasing amount of stress in our everyday working life. The causes of stress are diverse and often highly individual, so that sustainable stress management requires an individualised approach. The aim is to prove that a continuous multi-level assessment of stress is technologically feasible to a large extent with automated procedures and, combined with learning strategies, enables a personalised and sustainable stress management system for people in the work process.

This is achieved by means of a multimodal recording of stress symptoms on the various levels (physiological-physical, behavioural, emotional, cognitive), which is as far as possible automated and multimodal, supplemented by resources (personal, social, biological-physiological). A personalised user profile serves as the basis for a learning agent, which permanently optimises its knowledge base through sensor data and interactions with the user. Based on the user profile, the stress level and subjective and contextual information tailored to the current situation, the agent recommends personalised interventions for stress management. A neural network is used to determine the individual stress thresholds. The agent himself is trained over four phases with the aim of creating a personalised stress profile from which individual stress patterns can be derived using statistical methods for preventive stress management. The innovation of Relax lies in the modelling and translation of psychological concepts into technology, such as a neural network.

Relax is evaluated in three use cases reflecting different work contexts. A combination of several sensor data (pulse, heart rate, HRV, accelerometer, sleep recordings, skin conductance values on everyday objects) is used.