Improving EEG-based error detection using relative peak features

Abstract

A brain-computer interface (BCI) potentially enables a severely disabled person to communicate using brain signals. Automatic detection of error-related potentials (ErrPs) in electroencephalograph (EEG) could improve BCI performance by allowing to correct the erroneous action made by the machine. However, the current low accuracy in detecting ErrPs, particularly in some users, can reduce its potential benefits. The paper addresses this problem by proposing a novel relative peak feature (RPF) selection method to improve performance and accuracy for recognising an ErrP in the EEG. Using data collected from 29 participants with a mean age of 24.14 years the relative peak features yielded an average across all classifiers of 81.63% accuracy in detecting the erroneous events and an average 78.87 % accuracy in detecting the correct events, using KNN, SVM and LDA classifiers. In comparison to the temporal feature selection, there was a gain in performance in all classifiers of 17.85% for error accuracy and a reduction of -6.16% for correct accuracy Specifically; our proposed RPF used significantly reduced the number of features by 91.7% when compared with the state of the art temporal features.In the future, this work will improve the human-robot interaction by improving the accuracy of detecting errors that enable the BCI to correct any mistakes.