iSenseStress: assessing stress through human-smartphone interaction analysis

TY - GEN

T1 - iSenseStress

T2 - International Conference on Pervasive Computing Technologies for Healthcare 2015

AU - Ciman, Matteo

AU - Wac, Katarzyna

AU - Gaggi, Ombretta

N1 - Conference code: 9

PY - 2015/12/8

Y1 - 2015/12/8

N2 - Stress condition, if experienced for an extended amount of time, can negatively affect individual's health. Several external sensors monitoring different physiological states correlated with stress, or smartphone apps that monitor individuals context, have been leveraged to assess stress state in everyday life. The less intrusive human-smartphone interaction have been under-investigated so far. In our research we leverage 'swipe', 'scroll' and 'text input' interactions to assess the stress state of smartphone users. Based on data collected from 13 participants, we leverage 'swipe' and 'scroll' data to assess stress with an average F-measure of 79-85% for a within-subject model, and of 70-80% when building a global model. Moreover, 'text input' via a virtual keyboard has been analyzed, showing how several easy to calculate features enable to differentiate between stress and no-stress state. To the best of our knowledge, this is the first attempt to leverage human-smartphone interaction, and in particular 'swipe', 'scroll' and 'text input' interactions, to accurately assess stress state in individuals without using any external sensor or leveraging privacy-sensitive context information.

AB - Stress condition, if experienced for an extended amount of time, can negatively affect individual's health. Several external sensors monitoring different physiological states correlated with stress, or smartphone apps that monitor individuals context, have been leveraged to assess stress state in everyday life. The less intrusive human-smartphone interaction have been under-investigated so far. In our research we leverage 'swipe', 'scroll' and 'text input' interactions to assess the stress state of smartphone users. Based on data collected from 13 participants, we leverage 'swipe' and 'scroll' data to assess stress with an average F-measure of 79-85% for a within-subject model, and of 70-80% when building a global model. Moreover, 'text input' via a virtual keyboard has been analyzed, showing how several easy to calculate features enable to differentiate between stress and no-stress state. To the best of our knowledge, this is the first attempt to leverage human-smartphone interaction, and in particular 'swipe', 'scroll' and 'text input' interactions, to accurately assess stress state in individuals without using any external sensor or leveraging privacy-sensitive context information.

U2 - 10.4108/icst.pervasivehealth.2015.259280

DO - 10.4108/icst.pervasivehealth.2015.259280

M3 - Article in proceedings

SN - 978-1-63190-045-7

SP - 84

EP - 91

BT - 2015 9th International Conference on Pervasive Computing Technologies for Healthcare (PervasiveHealth)

PB - IEEE

Y2 - 20 May 2015 through 23 May 2015

ER -