FEATURE ANALYSIS OF FIVE FINGER MOVEMENT MOTOR IMAGERY ELECTROENCEPHALOGRAPHY SIGNAL

• Main Author: Hasanah, Syifa
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/56682
• Institution: Institut Teknologi Bandung

Brain-computer interface prosthetic hand is an endeavour to accelerate the quality of life among individuals with neuromuscular disease, amputation, and others. Prosthetic hand development under the BCI system has led to finger movement operation and intended to add convenience in daily activities. The study of feature extraction in BCI is limited to the central body part as feet and hands. Additionally, the previous research under the same dataset has not reached the minimum accuracy in operating BCI. The current project aims to increase the feature accuracy of motor imagery electroencephalography signal in five fingers classification. The differentiation performance analysis is conducted by exploring the features in various signal representations. Signal representation conversion began from the time-domain to the Fourier transform amplitude, power spectral density, and spectrogram. The signal feature of signal representation is computed based on channel-independent and channel-dependent. The majority of channelindependent feature calculation is confined to the statistical parameter. The analysis of signal representation as the signal feature is carried out in the project. The accuracy is observed further toward the frequency range. The classification is performed under the support vector machine in the manner of subject-dependent to quantify the differentiation performance. The highest accuracy is examined from many perspectives. The overall highest accuracy is 44.49% in spectrogram signal representation and a feature by contributing whole amplitude in transformation result. Due to its highest accuracy, further analysis on the evaluation metrics shows the highest specificity and sensitivity on the index (94%) and thumb (84%), respectively. The channel-independent of the statistical parameter has a score of 27.60% under the mean and sum signal feature. The channel-dependent has a 25.44% score under the signal feature related to the complement channel. The accuracy indicating the higher contribution for finger differentiation is a feature of the entire amplitudes of the transformation result. The frequency with the highest impact in differentiation ranges from 0 to 5 Hz.