Classification of Error-Related Potentials using Linear Discriminant Analysis

Error related potentials (ErrP) are evoked under various task situations and can be used for error correction. Here, we studied a specific ErrP, the so-called an observational ErrP. Six subjects participated in this study. They attempted the task of observing a moving cursor. The ErrP was elicited when the cursor moved in the opposite direction of the target instead of following it. We have employed linear discriminant analysis (LDA) classifier to classify both the correct and error events using only statistical means features. We have achieved a better average accuracy (81.43% in classifying correct events and 68.83% in classifying error events) than the Gaussian classifier. Also, we developed a generalized model using LDA to classify ErrP of subjects, whose data were not used in training. We have achieved the correct events classification accuracy, error events classification accuracy and area under the curve (AUC) 79.4%, 64.1%, and 0.777 respectively. The results revealed that by optimizing the generalized model, we can develop a one-fit for all model for detecting the error related potentials generated from other task situations.