Information Technology Reference
In-Depth Information
which begins to stop increasing. It is interesting that with semi-automatic approach,
accuracy is slightly better. This was also the same for Task1 vs. Relax. For Task2 vs.
Relax,
m
from semi-automatic approach gave the classification more stable in both
classes than the lower
m
as automatic gave (As we can see that classification of Task2
is slightly better, and it is the point that begins to stop increasing). Subject
g
, Task1
vs. Task2 showed successfully on our both selection approaches with same parameter
given as
m
= 3. The result is fine on Task1 vs. Relax. And for Task2 vs. Relax, auto-
matic approach gave the optimum value (
m
= 6), that classification of all aspect is
stable, with the best overall accuracy.
For dataset from BCI competition III; dataset IVa, subject
aa
showed that semi-
automatic selection provided the optimal value of
m
. While subject
al
was best at
m
=
1 which is the value given by the automatic approach. From the graph result of sub-
ject
av
, showed this subject is not a suitable user for using BCI application. As we can
see from the values around 3 to 11, those showed failure of classifier that failed for
discriminate Task 2, and overall accuracy is around 60% for all
m
values. Subject
aw
worked well with our both selection approaches. And for last subject, subject
ay
, our
both selection approaches give the optimum for overall accuracy.
7 Conclusion
Automatic selection and semi-automatic selection of prominent spatial filters of CSP
based on value of
m
provide an alternative method for BCI application when CSP is
used as a feature extraction tool. Automatic approach has advantage that the whole
process can be done by computer program without human intervention. However, in
some cases, it may not be the one that gives the highest accuracy. Semi-automatic
approach can alleviate from this drawback. Human can better interpret the relativity
between
r
2
values of each spatial filter and may select a better choice of
m
. However,
the selection can be subjective. The disadvantage of semi-automatic is of courses the
needs human to manually analyze the dropping rate, which may require some experi-
ence person to analyze them effectively. Analyzing of dropping rate may be hard to
determine by computer due to unpredictability on speed of dropping values for each
subject, and each pair of 2 classes. In our opinion, semi-automatic can be converted to
full automatic, if an efficient algorithm on analyzing of dropping rate was designed.
Acknowledgments.
We would like to thank Benjamin Blankertz, Guido Dornhege,
Matthias Krauledat, Klaus-Robert Müller, and Gabriel Curio for providing their data-
sets in our analysis.
References
1.
Koles, Z.J., Lazar, M.S., Zhou, S.Z.: Spatial Patterns Underlying Population Differences in
the Background EEG. Brain Topography 2(4), 275-284 (1990)
2.
Müller-Gerking, J., Pfurtscheller, G., Flyvbjerg, H.: Designing Optimal Spatial Filters for
Single-Trial EEG Classification in a Movement Task. Clinical Neurophysiology 110, 787-
798 (1999)
