Personalized acute stress classification from physiological signals with neural processes

Objective: A person's affective state has known relationships to physiological processes which can be measured by wearable sensors. However, while there are general trends those relationships can be person-specific. This work proposes using neural processes as a way to address individual differences. Methods: Stress classifiers built from classic machine learning models and from neural processes are compared on two datasets using leave-one-participant-out cross-validation. The neural processes models are contextualized on data from a brief period of a particular person's recording. Results: The neural processes models outperformed the standard machine learning models, and had the best performance when using periods of stress and baseline as context. Contextual points chosen from other participants led to lower performance. Conclusion: Neural processes can learn to adapt to person-specific physiological sensor data. There are a wide range of affective and medical applications for which this model could prove useful.