#TITLE_ALTERNATIVE#

• Main Author: HARYADI (NIM 23805003), BAGUS
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/10070
• Institution: Institut Teknologi Bandung

A brain computer interface is a communication system that does not depend on the brain's normal output pathways of peripheral nerves and muscles. One of BCI application in medical field is a promising technique to help people with severe motor disabilities to have effective control of devices such as computers, wheel chair or assistive appliances. BCI used a EEG as an instrument to measured of brain activity. One of problem in BCI research is to acquire the best recognition of classification from the input EEG signal. In this research, we are developed feature extraction which subject was asked to move a cursor up and down on a computer screen. Important element in BCI is EEG signal processing included feature extraction and feature classification.<p> <br /> <br /> <br /> <br /> <br /> In classification processes, input dimension of EEG data effect of computation time. So, we performed feature extraction to reduce of data without discard of main information of signal. On the feature extraction, power spectral and wavelet transformation with basis function of asymmetric symlet was used to extract feature of EEG signal. To classify feature of EEG signal that related with two classes features, we used artificial neural network based on Multi Layer Perceptron (MLP). To improve EEG signal classification, in the proposed techniques, on each channel of EEG data, MLP was trained using 268 sets of EEG data for each class in the learning step. Its procedure was repeated for other channels. On the testing step, two experiments were performed. First, 293 sets of EEG data used for single trial classification. Second, 290 sets of EEG data, the output from five trials classification was also compared with given of true class and the final class was determined by decision function.<p> <br /> <br /> <br /> <br /> <br /> Our results from classification with input feature-data from power spectral acquired 63 % correct. The best classification showed of input feature-data from wavelet transform using basis function symlet7 level 3 acquired 82% of total tested data from channel A1-Cz. The improvement method with five trial classification shows that the correct classification can be increased up to 97%.