Predicting occurrence of errors during a Go/No-Go task from EEG signals using Support Vector Machine

Human error often becomes a serious problem in dairy life. Recent studies have shown that failures of attention and motor errors can be captured before they actually occur in the alpha, theta, and beta-band powers of electroencephalograms (EEGs), suggesting the possibility that errors in motor responses can be predicted. The goal of this study was to use single-trial offline classification to examine how accurately EEG signals recorded before motor responses can predict subsequent errors. Ten subjects performed a Go/No-Go task, and the accuracy of error classification by a Support Vector Machine (SVM) was investigated 1000 ms before presenting the Go/No-Go cue. The resulting mean classification accuracy was 62%, and strong increases and decreases in activities associated with errors were observed in occipital and frontal alpha-band powers. This result suggests the possibility that future errors can be predicted using EEG.