Page - 157 - in Intelligent Environments 2019 - Workshop Proceedings of the 15th International Conference on Intelligent Environments
Image of the Page - 157 -
Text of the Page - 157 -
2.2.4 Creating feature quantities
The EEG of 60 seconds was divided every 6 seconds, each segment was divided into
three frequency bands, and the energy of each frequency band was calculated. Feature
quantity was created based on this energy. The created feature quantity is the average
value, the median value, the variance, the maximum value, and the minimum value of
the ten divided segments. Since these five types of feature quantities exist for three
frequency bands, 15 feature quantities were created from the original EEG of 60 seconds.
Specifically, the average value of γ wave, the median value of γ wave, γ wave dispersion,
γ wave maximum value, γ wave minimum value, β wave average value, β wave median
value, β wave dispersion, β wave maximum value, β wave minimum Value, the α wave
maximum value, the α wave minimum value, the α wave mean value, the α wave median
value, the α wave dispersion, the α wave maximum value, and the α wave minimum
value. We analyzed based on these 15 features. Originally, the frequency band should
not be divided into three, but rather it should be separated more finely and the type of the
feature quantity should be increased to carry out more detailed analysis, but as the feature
quantity increases, the analysis becomes complicated, we considered that we should
increase the amount of features after establishing analytical methods, so we used 15
feature quantities in this study.
By the above processing, a label of D defined in eqn.1 was newly created from the
labels of Valence and Arousal that were attached to the original data set. In the EEG,
sign also obtained from each electrode site were separated into three frequency bands α,
β, and γ, and feature quantities were created based on the calculated energy.
2.3. Investigation by LARS
LARS is an algorithm to calculate the Lasso estimated value. Lasso can make variable
selection as to which objective variable is important for the objective variable. LARS
is mainly used to prevent over learning or to select variables. We thought that it would
be possible to know the electrode with a high contribution to D by investigating the
feature amount with high contribution to LARS for the objective variable D. In fact, by
changing the number of features from 1 to 5, the average value of energy in each
frequency band was calculated by LARS(eqn. 4) [17].
c(̂) = X
’
(y − ̂) = X ’
( − ̂) (4)
Table 2 . Result of feature quantity selection using LARS
Number of
features Features
1 F4 Median
2 F4 Median AF4 minimum value
3 F4 Median AF4 minimum value
T7 Mean
4 F4 Median AF4 minimum value
T7 Mean Pz Maximum value
5 F4 Median AF4 minimum value
T7 Mean Pz Maximum value
FC6 Median
Y.Tsurugasaki etal. / IdentificationofEffectiveEEGElectrodes forDepressionSensing 157
Intelligent Environments 2019
Workshop Proceedings of the 15th International Conference on Intelligent Environments
- Title
- Intelligent Environments 2019
- Subtitle
- Workshop Proceedings of the 15th International Conference on Intelligent Environments
- Authors
- Andrés Muñoz
- Sofia Ouhbi
- Wolfgang Minker
- Loubna Echabbi
- Miguel Navarro-Cía
- Publisher
- IOS Press BV
- Date
- 2019
- Language
- German
- License
- CC BY-NC 4.0
- ISBN
- 978-1-61499-983-6
- Size
- 16.0 x 24.0 cm
- Pages
- 416
- Category
- Tagungsbände