Feller, Shani2021-05-052021-05-052020https://hdl.handle.net/10539/31108A dissertation submitted to the Faculty of Engineering and the Built Environment, University of Witwatersrand, Johannesburg, in fulfilment of the requirements for the degree of Master of Science in Engineering, 2020A robotic/orthotic hand that is controlled by an electroencephalograph (EEG)-based brain computer interface (BCI) which can perform wrist extension (WE) and wrist flexion (WF), can benefit motor impaired individuals to perform everyday activities. Numerous EEG electrodes are typically needed for the spatial resolution required for this task. However, this makes the system expensive, time-consuming to set up and uncomfortable for the user. This research explores the development of a new electrode reduction method that produces an optimised and reduced EEG electrode set, when differentiating between WE and WF movements. A 128-electrode, EEG database consisting of recordings from 5 subjects was used in this study. The method utilises independent component analysis (ICA) and features related to event-related desynchronization and resynchronization (ERD/ERS) modulations, to produce a reduced electrode set. The method was applied to two movement type investigations: 1) right-hand and left-hand movements and 2) unilateral WE and WE movements. The effectiveness of the method was tested by evaluating the amount of motor-control information that was retained during the electrode reduction. This was done by combining the analysis of three tests: 1) classification accuracy. 2) visual comparison of the inter-trial variance plots of ERD/ERS. 3) comparing the amount of variance retained across all electrodes sets. The results suggest that the optimal channel configuration for both investigations was a 16-electrode configuration. The 16-electrode configuration obtained a classification accuracy of 70.51 % and 61.61 % for investigation 1 and investigation 2 respectively which is a loss of classification accuracy of 12.01% and 11.16 % respectively, when compared to using the full electrode set. This research illustrates the potential in using ICA as a primary technique for electrode reduction and suggests that non-motor related electrodes may have an important role in recording motor-related cortical activityenThesis