Detecting P300 Potentials Using Weighted Ensemble Learning

The P300 is an event-related potential seen in electroencephalogram (EEG) waveforms. It is elicited by visual stimuli and used by a brain-computer interface (BCI) known as the P300 speller for character input. However, EEG traces are generally very noisy and can bury the characteristic component in the noise, so various feature extraction methods have been used to extract such features from EEG results. In this paper, we investigate the discriminating P300 potentials using a convolutional neural network (CNN), a widely-used image processing tool. We also consider weighted ensemble learning combining a CNN, stepwise linear discriminant analysis (SWLDA), and a support vector machine (SVM), with the aim of improving the P300 detection accuracy.