Brain Computer Interface (BCI) is getting a lot of attention from researchers because
BCI is a system used to translate, manage and recognize human brain activity.
Electroencephalography (EEG) is one type of BCI which is included in non-invasive
because EEG uses external sensors to measure brain activity. However, EEG has a
non-stationary characteristic. Therefore the information on the EEG signal is difficult
to process.
This thesis proposes (i) converting the EEG signal into an image and (ii) optimizing
the BCI system using feature selection and channel selection. The data used
is the EEG stroke signal data set from Universiti Teknologi Malaysia. EEG signal
feature extraction with the power spectrum density (PSD). The value of the energy
distribution is carried out by brain mapping for each channel. The image feature
extraction used is GLCM with 11 statistical characteristics. Feature selection using
the GA, MI, and Chi-Square methods to find the optimal method for the system.
Therefore, in this thesis, we propose channel selection in the energy distribution
image to eliminate irrelevant channels by taking the value of the selected channel
on the channel. The classification method in this thesis is Artificial Neural Network
Back-Propagation (ANN-BP). The validity of the proposed BCI system is training
accuracy, test accuracy, time complexity, and brain mapping.
The feature selection results using the GA method are 7 GLCM characteristics:
correlation, energy, homogeneity, inverse difference momentum, different variance,
and sum variance. Optimization with channel selection produces 9 channels: AF4,
FCz, FC4, FT8, C4, T8, Cp4, Pz, and Oz. Compared with the image system without
and with channel selection, the accuracy with channel selection can improve 5%
and the time complexity is faster, with a gap of 1,928 ms. Since the EEG signal
has a non-stationary characteristic that makes each class challenging to identify,
irrelevant values can be omitted because they can confuse the system. This thesis
generates the optimal system by using an energy distribution image system using
the GA-GLCM feature selection and channel selection.
Keywords: EEG signal, energy distribution image, channel selection, feature
selection, brain mapping.