Multi-Objective Binary-PSO for Improving the Performance of P300 Speller

P300 speller provides a medium of communication to patients with severe neuro-muscular disabilities. It relies on the detection of P300 peak in EEG recordings due to the occurrence of external stimulus. The patients are asked to focus their attention on the character/symbol displayed on the screen and their EEG recordings are monitored at the same time. These recorded signals are then processed and analyzed to determine the target character. Thus the patient is able to spell out the word he intended to communicate. In the process, a large amount of redundant and noisy data is also collected which reduces the overall accuracy of the speller. This paper proposes the use of multi-objective binary particle swarm optimization (MO-BPSO) technique to identify the most relevant electrode channels on an EEG headset for data collection. The trade-off between the accuracy of the character detection and the number of channels is determined using the pareto-optimal solution plot. The user can later select any of the pareto-optimal solution based on his requirement. The dataset used for this purpose is of BCI competition III. The result section also discusses the channel and the positioning which play a major role in P300 signal detection.