EEG Data Analysis in HMI Representation

Ibrahim Salem Jahan

Doctoral thesis

Abstract:

Nowadays, with the progress of science and technology in the field of signal analysis, data analysis and data mining are becoming very significant factors in science and engineering applications. Extracting useful knowledge from experimental raw datasets, measurements, observations and analysis, and understanding complex data have all become global matters of interest. Raw datasets, most commonly collected from complex phenomena, either express integrated results of several hidden, related variables, or they are a set of underlying, hidden components of factors. The complex raw dataset first must be decomposed by a dimensional reduction method, i.e. matrix decomposition, to extract hidden information or hidden factors of a complex raw dataset. One very complex and high-dimension data type is the EEG signal. EEG data has several applications (e.g. diagnosis of brain disease) that can be used to improve control devices to better aid the handicapped in interacting with their surroundings. Once we have successfully analyzed and classified EEG signals, we can assign each mental task to a unique control command. For this purpose, and within the scope of our work, we have employed several techniques, i.e. Faster Fourier Transform (FFT), Polynomial Curve Fitting, Turtle Graphics, LZ complexity, and a Self Organizing Map (SOM) neural network. In the future, we plan to also employ other Dynamic Time Warping (DTW) technique. We have combined these techniques to recognize and classify either EEG signals or mental tasks. We have carried out some experiments on EEG data. The first experiment was on EEG data for recognition of EEG hand movement signals, while the second and third experiments focused on the detection of index finger movement. A fourth experiment, for the classification of EEG data based on Dynamic Time Warping (DTW), is in progress. Our maximum classification accuracy rate in the first three experiments reached 96%. Results for the fourth experiment are pending, however, results from our previous experiments have shown improved accuracy rates when compared to other methods. Our conclusion also addresses comparisons between our method and other existing methods.

Abstract:

Nowadays, with the progress of science and technology in the field of signal analysis, data analysis and data mining are becoming very significant factors in science and engineering applications. Extracting useful knowledge from experimental raw datasets, measurements, observations and analysis, and understanding complex data have all become global matters of interest. Raw datasets, most commonly collected from complex phenomena, either express integrated results of several hidden, related variables, or they are a set of underlying, hidden components of factors. The complex raw dataset first must be decomposed by a dimensional reduction method, i.e. matrix decomposition, to extract hidden information or hidden factors of a complex raw dataset. One very complex and high-dimension data type is the EEG signal. EEG data has several applications (e.g. diagnosis of brain disease) that can be used to improve control devices to better aid the handicapped in interacting with their surroundings. Once we have successfully analyzed and classified EEG signals, we can assign each mental task to a unique control command. For this purpose, and within the scope of our work, we have employed several techniques, i.e. Faster Fourier Transform (FFT), Polynomial Curve Fitting, Turtle Graphics, LZ complexity, and a Self Organizing Map (SOM) neural network. In the future, we plan to also employ other Dynamic Time Warping (DTW) technique. We have combined these techniques to recognize and classify either EEG signals or mental tasks. We have carried out some experiments on EEG data. The first experiment was on EEG data for recognition of EEG hand movement signals, while the second and third experiments focused on the detection of index finger movement. A fourth experiment, for the classification of EEG data based on Dynamic Time Warping (DTW), is in progress. Our maximum classification accuracy rate in the first three experiments reached 96%. Results for the fourth experiment are pending, however, results from our previous experiments have shown improved accuracy rates when compared to other methods. Our conclusion also addresses comparisons between our method and other existing methods.

Keywords

Electroencephalography (EEG), EEG Data, EEG Classification, EEG Pattern Recognition, Feature Extraction, LZ complexity, Self Organizing Map, Dynamic Time Warping.

Keywords

Electroencephalography (EEG), EEG Data, EEG Classification, EEG Pattern Recognition, Feature Extraction, LZ complexity, Self Organizing Map, Dynamic Time Warping.

Language used: English

Date on which the thesis was submitted / produced: 25. 2. 2015

Thesis defence

• Date of defence: 11. 6. 2015
• Supervisor: Václav Snášel
• Reader: Petr Čermák, Marek Penhaker, Roman Šenkeřík

VŠB - Technical University of Ostrava

Fakulta elektrotechniky a informatiky

Doctoral programme / field:
Informatika, komunikační technologie a aplikovaná matematika / Informatika