An Offline Study for a Single-Trial ERP Card-Guessing Game

Abstract

Event-Related Potential (ERP) is one of the brain signal features which are used in Electroencephalography (EEG) based research and application development, such as Brain-Computer Interface (BCI) applications. Recently, single-trial ERP has been one of the main interests of researchers in the field of BCI and neuroscience. In this paper, an offline study which evaluated the feasibility of developing an online BCI guessing game, based on single-trial ERP, was presented. The objective was to determine the optimal methods and parameters needed to achieve high online classification accuracy and performance. Eight subjects participated in our experiments to collect the data for the offline study. Each subject had to choose one out of six cards displayed on a computer monitor. Three different algorithms of Linear Discriminant Analysis (LDA) were used for classifying the cards into targets and non-targets. Canonical Correlation Analysis (CCA) was applied as a spatial filter for the 16-channel data. Additionally, the data were analysed and classified per channel to deduce which channel reached the higher performance. The results proved the feasibility of the online application. The best performance was achieved with the personalised data and by taking the majority vote of the three LDA algorithms.