Single-trial EEG classification of similar errors

When humans recognise errors, either committed by themselves or observed, error-related potentials (ErrP) are produced in the brain. Recently, a few studies have shown that it is possible to differentiate between the ErrPs generated for errors of different direction, severity, or type (e.g., response errors, interaction errors). However, in real-world scenarios, errors cannot always be delineated by these metrics. As such, it is important to consider whether errors that are similar in all of the aforementioned aspects can be classified against each other on a single-trial basis. In this paper, for the first time, we consider two different response errors, which are of equal severity and have no associated direction. This study used electroencephalogram (EEG) data from a sustainedattention based time-critical reaction task, where time pressure caused subjects to commit two different errors. Using data from 16 subjects, we applied time domain EEG features and an ensemble of linear classifiers to separate these two error conditions on a single-trial basis. We achieved a mean balanced accuracy of 63.23% and, for most of these subjects, achieved statistically significant (p <; 0.05) separation of the two error conditions. The ability to classify similar error conditions, such as these, increases the scope of possible applications for EEG error detection, and has the potential to improve brain-machine interaction.