#TITLE_ALTERNATIVE#

• Main Author: SYIRFASARI (NIM 13204172), PUTRI
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/11177
• Institution: Institut Teknologi Bandung

Brain-Computer Interface is one kind of technology that can help the interaction between human and computer which is based on EEG signal processing. The main purpose of EEG signal processing is to translate brain signal that is generated when people are doing certain action to computer language.<p> <br /> <br /> <br /> <br /> <br /> This research used self-paced key typing data which has already been free from artifacts. Self-paced key typing data contained information of EEG signal that was recorded when someone was typing using keyboard with certain condition. Processes that have to be done in the next step are feature extraction and feature classification.<p> <br /> <br /> <br /> <br /> <br /> Before feature extraction using FFT and PCA, pre-processing should be done to simplify the next computation. The pre-processing that has been done was data separation based on channel and class. After this pre-processing, the data went to FFT. It gave some result but only the first eight coefficient of the FFT result was taken and put to PCA. The first two components, which have the biggest variance, would be the features.<p> <br /> <br /> <br /> <br /> <br /> After feature extraction, features were put into the classifier to recognize the feature pattern to each class, class 'left' and class 'right'. Some features belong to class 'right' and some features belong to class 'left'. There were two classifiers that have been used in this research, Nave Bayesian Classifier and Neural Network-Back Propagation.<p> <br /> <br /> <br /> <br /> <br /> After all the feature recognition has been done, test of system should be done by using test data in order to give the label to each test data. The success of the system can be seen by the success of the system giving the exact label to each data comparing to the true label.<p> <br /> <br /> <br /> <br /> <br /> By using Nave Bayesian Classifier, it did not give the good result. Almost every test data got the label for class 'left' because class 'left' had bigger probability in training data and Gaussian distribution for each feature was not different to each other. Comparing to the true label, then its error would be 51%.<p> <br /> <br /> <br /> <br /> <br /> By using Neural Network-Back Propagation, it gave a good result. This classifier could do the classification based on the training that has been done before. The error would be 28%.