Noninvasive stress recognition considering the current activity

In this study, we aim to determine the stress level in a non-invasive way to the maximum extent possible by analyzing behavioral and contextual data received from the only source being a smartphone containing the data gathered in real-life situations. The information collected includes audio, gyroscope and accelerometer features, light condition, screen mode (on/off), current stress level self-assessment, and the current activity type. Three stress analysis models have been built: two with the consideration of current activities of a participant and one without those. Classification of low- and high-stress conditions, which was executed for a separate model for a certain kind of activity only, enabled us to achieve 3.9 % higher accuracy than that under the conditions when those activities were neglected. Also, the Android application was developed as a means for the current activity-type identification.