Detecting Typing Errors in a Numerical Typing Task with Linear Discriminant Analysis of Single Trial EEG

Abstract

Linear Discriminant Analysis (LDA) was utilized to detect numerical typing errors in the context of daily data input. Single EEG trial data from 6 subjects were analyzed and a 67% detection rate was demonstrated by Fisher LDA classifier with an optimal Mahalanobis distance ratio setting. Sensitivity analysis showed that Fisher LDA classifier detected the errors in terms of 9-digit numbers by 62.19% on average, in comparison with 3.33% and 47.22% using the prior model and the chance model. This is one step towards predicting human errors in perceptual-motor tasks before their occurrence; future work would focus on benchmarking to improve current method toward an online and robust classifier.