Generic pattern recognition models based on EEG-MI brain computer interfaces for wheelchair steering control
The purpose of this study was to develop generic pattern recognition models (GPRMs)based on two-class EEGMI brain-computer interfaces for wheelchair steering control.Initially, a preprocessing procedure was performed to remove unwanted signals and toidentify the optimal duration of MI feature compon...
Saved in:
| Main Author: | |
|---|---|
| Format: | thesis |
| Language: | eng |
| Published: |
2020
|
| Subjects: | |
| Online Access: | https://ir.upsi.edu.my/detailsg.php?det=6509 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The purpose of this study was to develop generic pattern recognition models (GPRMs)based on two-class EEGMI brain-computer interfaces for wheelchair steering control.Initially, a preprocessing procedure was performed to remove unwanted signals and toidentify the optimal duration of MI feature components. Then, feature extraction basedon five statistical features, namely min, max, mean, median, and standard deviation wereutilized for extracting the MI feature components in three signal domains, namely time,frequency, and time-frequency domains. Seven classification algorithms, namely LDA,SVM, KNN, ANN, NB, DT, and LR were selected and tested to find the best algorithmsthat could be used for the development of hybrid classifiers. Two datasets were used,namely the BCI Competition dataset (which belonged to Graz University) and theEmotive EPOC dataset (which was collected in this study), with the former beingutilized in the development, evaluation, and validation of the GPRM models and thelatter being used for validation only. The research findings showed that GPRM modelsbased on the LR classifier were highly accurate in the time and time-frequency domainsin the range of 4 and 6 seconds and 4 and 7 seconds, respectively. In addition, GPRMmodels based on the MLP-LR classifier were highly accurate in the frequency domainin the range of 4 and 6 seconds. Furthermore, the validation of such models using theEmotive EPOC dataset showed that the LR-based GPRM model attained highclassification accuracies of 90.2% and 85.7% in the time domain and time-frequencydomain, respectively. The MLP-LR-based GPRM models achieved a classificationaccuracy of 84.2% in the frequency domain. In conclusion, the main findings showedthat GPRMs were highly adaptable when deployed in the real-time application of theEEG-MI-based wheelchair steering control system. The implication of this study is thatgeneric pattern recognition models based on EEG-MI Brain-Computer interfaces can beutilized to improve the effectiveness of wheelchair steering control. |
|---|