Information Technology Reference
In-Depth Information
9.1.2 Main Contributions
The aim of the work presented in this chapter was to construct a model that is able
to predict the alertness state of a human using one electrode; and this model will be
used in real time applications. That is why, the two main objectives are:
Reduce the time needed to install the EEG cap on a participant using a variable
selection method in order to choose the best electrode (based on classi
cation
rate). In fact, in real-world applications, it is necessary to reduce the number of
electrodes needed because the cap installation process has to be short. A long
installation of the EEG cap can cause a disturbance of the mental state of the
person that we want to study (pilots or surgeons for example).
To obtain a model (decision rule) that is able to give a reliable prediction of the
alertness state of a new participant.
To achieve these objectives, we apply a wavelet decomposition as a preprocessing
step and a new criterion for state discrimination is proposed. Then, several standard
methods for supervised classi
cation (binary decision tree, random forests, and
others) are used to predict the state of alertness of the participants. The criterion is then
re
ned using a genetic algorithm (GA) to improve the quality of the prediction.
Finally, this work presents results that are part of a broader research program that is
being investigated by the lead authors, focusing on the development of BCIs. In
particular, this chapter contains a detailed description of the system originally pre-
sented in V
zard et al. (
2014
), where critical aspects were not discussed in detail.
The remainder of this chapter proceeds as follows. The data acquisition protocol is
precisely detailed in the Sect.
9.2
. The validation of the data is described in the
Sect.
9.3
. A data preprocessing is proposed in Sect.
9.4
and a feature extraction is
performed Sect.
9.5
in order to compute a
é
cation of EEG signals.
Section
9.6
contains the general principles of a GA and presents how this stochastic
optimization method improves the results obtained in the previous section. Finally,
Sect.
9.7
presents a summary of this work and discusses our main conclusions.
rst attempt of classi
9.2
Data Acquisition
This work is based on real data that we have collected. This section will describe
the data acquisition and data validation steps.
9.2.1 Participants
This work uses 44 participants, with ages between 18 and 35, all are right-handed,
to avoid variations in the characteristics of the EEG due to age or handedness linked
to a functional interhemispheric asymmetry.
