Information Technology Reference
In-Depth Information
electrodermal activity, another name for Galvanic
skin response (GSR). Four different emotions were
induced: sadness, anger, stress and surprise by
reading a story with both the voice and the story
reflecting the emotion induced. Then support vec-
tor machine for regression is used for classification
and a 67.5% of correct-classification ratio was
obtained (Kim, et al., 2004). The method can be
improved by using multi-class SVM or by using
one of its variants. Other work proposed detect-
ing human emotion using only the ECG. From
the ECG, three features were extracted: energy,
recoursing energy efficiency and root mean square.
Four emotions were induced using audiovisual
methods: happiness, disgust, surprise and fear.
Classification was done using unsupervised fuzzy
Cmean clustering where db4 wavelet transform
was used for feature extraction. No accuracy was
reported but the graphs show a good accuracy of
clustering (Murugappan, et al., 2008). Fisher pro-
jection was used with sequential floating forward
search to achieve an accuracy of 81% in detecting
8 classes of emotion including neutral. This study
has proven the feasibility of emotion detection and
that the features extracted for all the emotions on
a specific day are clustered more tightly than the
features of a specific emotion during multiple
days (Picard, et al., 2001).
A high percentage of the agitation manage-
ment techniques require efficient and accurate
agitation detection. Although not all for patients
with dementia, many different agitation detection
techniques were previously developed. Some
techniques use hidden Markov models to estimate
the inputs to an SVM, using an outside camera to
observe the movement of the patient while sleep-
ing (Fook, et al., 2007). The limitations of this
approach are privacy concerns and the inability
to monitor the patient all day. Another limitation
is in the Markov model used. It is known that
any model is called a Markov model if: knowing
the present, the future is independent of the past.
But for agitation detection, knowing the present
is not always enough to predict the future and
historical behavior must be known. Some other
techniques of frustration detection use facial
expression, head movement and eye movement.
A dynamic Bayesian network model was used to
integrate all these features as well as GSR, RTD
and Blood Volume Pressure (BVP) (Wenhui, et
al., 2005). The user must be sitting in front of a
computer and holding the mouse to be able to
measure all the inputs which is a limitation for
elderly subjects and cannot be clinically used. A
theoretical limitation to this method resides in
the fact that using a Bayesian model means that
the underlying probability densities between the
output of the system and its inputs are known.
But this probability density is in most cases not
known and cannot even be estimated. Hence, the
need for a learning algorithm, such as SVM, that
does not require the underlying probability densi-
ties. Some studies done on the acceleration of the
wrists, ankles, and waist found that agitated people
had sudden movements of the wrists and ankles
(Tamura, et al., 1997). The most relevant study
that was done so far was based on the measure-
ment of BVP, GSR, ST and the pupil diameter.
From the heart rate they extracted the Inter Beat
Interval (IBI) and then studied it in the frequency
domain because the ratio of the low frequency
component over the high frequency component
could be an indicator of stress (Zhai, et al., 2006).
Eleven features were extracted from these sensors
and then fed into an SVM. The elicitation of stress
was done using the Stroop test (Stroop, 1935). The
SVM was tested using the cross validation tech-
nique and results showed 90.1% accuracy (Zhai,
et al., 2006). The limitation of this technique is
the use of 11 features that are fed into the SVM
which is computationally demanding for a micro
controller implementation. Moreover, the user has
to sit in-front of a computer to be able to capture
the pupil diameter. Previous work done by our
team has shown that subject independent agitation
detection was possible using SVM, by monitoring
three vital signs: heart rate (inter-beat interval),
Galvanic skin response and skin temperature. It
Search WWH ::
Custom Search