iSenseStress: Assessing Stress Through Human-Smartphone Interaction Analysis Matteo Ciman 1 , Katarzyna Wac 2 and Ombretta Gaggi 1 1 University of Padua Padua, Italy 2 University of Geneva and University of Copenhagen Geneva, Switzerland and Copenhagen, Denmark
27
Embed
iSenseStress: Assessing Stress Through Human-Smartphone Interaction Analysis
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
iSenseStress: Assessing Stress Through Human-Smartphone
Interaction AnalysisMatteo Ciman1, Katarzyna Wac2 and Ombretta Gaggi1
1 University of Padua Padua, Italy
2 University of Geneva and University of Copenhagen Geneva, Switzerland and Copenhagen, Denmark
PervasiveHealth 2015, Istanbul, Turkey /26
Stress Experience
• Stress is mental condition experienced every day
• Long exposure can lead to anxiety, depression etc. => increase of healthcare costs
• In 2013 American teens reported stress experienced at unhealthy levels (and at increasing lower ages) [http://www.apa.org/news/press/releases/stress/2013/teenstress.aspx]
2
• Early assessment of stress condition can help to provide feedback to improve health state of individuals
[1] D. Sun, P. Paredes, and J. Canny, “Moustress: Detecting stress from mouse motion,” in SIGCHI Conference on Human Factors in Computing Systems, 2014, pp. 61–70. [2] G. Bauer and P. Lukowicz, “Can smartphones detect stress-related changes in the behaviour of individuals?” in PERCOM Workshops, 2012.
PervasiveHealth 2015, Istanbul, Turkey /26
The idea
4
PervasiveHealth 2015, Istanbul, Turkey /26
Our Approach
• No external devices used, just smartphone (less expensive, more usable)
• No privacy-related information (i.e., calls, messages, location etc.)
• Possible to run a phone background service all the day long
• Based on human-smartphone interaction analysis
• Limitation: an interaction with the smartphone is required to make an assessment
5
PervasiveHealth 2015, Istanbul, Turkey /26
Human-Smartphone Interaction
6
Tap
Scroll
Swipe
Text WritingDouble
Tap
Rotate
Zoom
Pinch
Long press
PervasiveHealth 2015, Istanbul, Turkey /26
Tasks Definition
• Search Task:
• Scroll, swipe and tap
• Write Task:
• Tap, Text Writing
7
Tap Scroll Swipe
Tap Text Writing
PervasiveHealth 2015, Istanbul, Turkey /26
Search Task
• Find inside a 21x15 grid the right icon (s)
• Scroll and Swipe to inspect all the icons
• Tap to select the right icon
8
PervasiveHealth 2015, Istanbul, Turkey /26
Search Task Features
• Tap {min, max, average} pressure / length / size
• Statistical analysis for significance evaluation
• Stress prediction model using Decision Tree (DT), k-Nearest Neighbourhood (kNN), Bayes Network (BN), Support Vector Machine (SVM) and Neural Networks (NN)
• User and global model (evaluated using 10-Fold cross validation and leave-one-out)
21
PervasiveHealth 2015, Istanbul, Turkey /26
Search Task - Statistical Correlation• Only weak correlation between our features
• Global Model
• Average swipe pressure (p-value = 0,09)
• Scroll distance from center (p-value = 0,065)
• Scroll distance from top left (p-value = 0,07)
22
• User model
• Scroll interaction length (strong correlation for 61% of users)
• Scroll delta (strong correlation for 40% of users)
• Scroll linearity (strong correlation for 45% of users)
PervasiveHealth 2015, Istanbul, Turkey /26
Search Task - Prediction Model
23
F-measure for Scroll interaction models
MODEL DT KNN SVM NN BN
USER (AVERAGE) 0.79 0.80 0.81 0.80 0.77
GLOBAL (AVERAGE) 0.73 0.71 0.78 0.74 0.67
F-measure for Swipe interaction modelsMODEL DT KNN SVM NN BN
USER (AVERAGE) 0.86 0.86 0.79 0.87 0.85
GLOBAL (AVERAGE) 0.92 0.75 0.81 0.82 0.77
PervasiveHealth 2015, Istanbul, Turkey /26
Write Task - Statistical Correlation
• User Model
• Digits size (64% of users with strong correlation)
• Pressure/Size ratio (55% of users with strong correlation)
• Global Model
• Wrong Words / Total words ratio (p-value = 0,028)
• Digits time distance (p-value = 0,012)
• Digit duration (p-value = 0,08)
24
PervasiveHealth 2015, Istanbul, Turkey /26
Conclusions• Stress assessment using data from non-intrusive
devices can increase people’ acceptance
• Human-smartphone interaction analysis can be leveraged to assess stress state in users
• Scroll and Swipe: F-measure of stress prediction between 79% and 85% for user models, and between 70% and 80% for global model.
• Text writing: several features showed strong correlation
25
PervasiveHealth 2015, Istanbul, Turkey /26
Future works
• Real-time background service for stress assessment
• Behaviour suggestion implementation
• Stress assessment in the wild (ongoing study, 29 participants)
26
• My PhD Thesis :)
iSenseStress: Assessing Stress Through Human-Smartphone
Interaction AnalysisMatteo Ciman, Katarzyna Wac and Ombretta Gaggi