Document 6603388

Using Electrodermal Activity to Recognize Ease of
Engagement in Children during Social Interactions
Javier
Work supported by the National Science Foundation
Grant No. NSF CCF-1029585
1
Hernandez ,
Ivan
2
Riobo ,
Agata
Experimental Setting
Hello!
Children’s emotional self-regulation and coregulation are key components in understanding
engagement.
Ball
Book
Hat
Tickle
Affectiva QTM sensor
Characterization of EDA Responses
0.55
EDA levels stable
Shorter sessions
(N: 29)
0.65
0.6
0.55
0.5
0.5
0.4
0.35
0.25
0
40
20
60
80
100
Avg. Time (seconds)
120
0.35
0.3
0
(Responses were resampled to last the average session time)
(Benedek and Kaembach, 2010)
20
40
60
80
100
Avg. Time (seconds)
120
160
140
Individual Features (IF); only from child’s EDA
Standard
Deviation
# Peaks
Average Peak
Amplitude
Mean
Slope
Position
Minimum
Area under the curve
Position
Maximum
Synchrony Features (SF); from the dyad’s EDA
Normalized EDA
0.75
0.5
0.25
Child
Adult
Tonic/Phasic
Components
Tonic
Phasic
0
0:56
0.4
0.25
140
Processing
1. Normalize EDA values
2. Reduce noise
3. Extract tonic and phasic EDA
0:00
0.45
g
EDA levels increase
Longer sessions
(N: 22)
0.45
0.3
1
0.7
1:12
Time (minutes:seconds)
1:68
2:34
Time
Correlation between responses:
• Pearson correlation
• Canonical Correlation
• Dynamic Time Warping
Differences between features:
• Means
• # Peaks
• Avg. peak amplitude
Average AUCs
(ROCs and Precision/Recall)
0.6
Limitations
• Specificity
• Artifacts
• Wireless
• Comfortable
• 32 Hz
• 4 sensors
Easier vs Harder to Engage with SVMs
Harder to Engage Children
0.75
Tonic/Phasic
Components
Avg. Normalized EDA
(±Standard Error)
0.65
Good Indicator
• Arousal
• Cognitive Load
23 sessions excluded (artifacts)
51 sessions used for analysis
• Easier to engage (N: 29)
• Harder to engage (N: 22)
Ratings per activity:
• 0 – little effort
• 1 – some effort
• 2 – extensive effort
0.7
Rosalind W.
1
Picard
Electrodermal Activity (EDA)
23 sessions exclude due to artifacts
51 clean sessions
Engagement groups
External coder
• Easier to engage (N = 29)
“Amount of effort required to get child’s attention”
• Harder to engage (N = 22)
Can we characterize qualitative aspects of
children’s social engagement with wearable
biosensors?
Easier to Engage Children
Gregory D.
2
Abowd ,
{javierhr, picard}@media.mit.edu – MIT Media Lab1 - {ivan.riobo, agata, abowd}@gatech.edu – Georgia Tech2
Motivation
0.75
2
Rozga ,
90
85
Combining Features with
Sequential Forward Selection
80
75
70
65
60
55
50
Top
feature
Top
feature
Diff.
#peaks
of dyad
from
phasic
Position
max.
from
tonic
IF+SF
IF
One Feature
at a Time
Top
feature
DTW
from
phasic
Diff.
#peaks
of dyad
from
phasic
Position
max.
from
tonic
SF
IF
SF
SF and IF similar performance
IF better from tonic
SF better from phasic
Tonic and Phasic decomposition improved >6%
Feature selection improved >11%
SF slightly better than IF (>2%)
Tonic and phasic equally represented
SF and IF are complementary
Hernandez J., Riobo I., Rozga A., Abowd G. D., Picard R. W., (2014), Using
Electrodermal Activity to Recognize Ease of Engagement in Children during
Social Interactions, In Proc. of Inter. Conf. on Ubiquitous Computing, 307-17