Identification of Error-Related Observations from Event Related Potentials Using Pattern Recognition Techniques

This article presents a comparison of pattern recognition techniques that have been used for the identification of correct or incorrect actions by means of Event Related Potentials (ERPs). ERP data from 47 electrodes were acquired from sixteen volunteers (observers), who observed correct or incorrect responses of subjects (actors) performing a special designed task. First and second order statistical features, features from the frequency domain as well as the Σ-Φ-Ω features were extracted from the ERP signals. Three different feature selection methods were applied, namely the Sequential Floating Forward Selection (SFFS), the Wilcoxon rank testing and the Genetic Algorithms. Two supervised algorithms, Artificial Neural Networks (ANN) and Support Vector Machines (SVM), and one unsupervised algorithm, Fuzzy C-Means (FCM), were used for the classification task. Five combinations of feature selection and classification algorithm were compared with respect to their classification accuracy by means of the leave one out method. Results indicate that classification accuracy between 84.4% and 100% can be achieved. Introduction A significant part of the learning process in human development takes place through observation. The behavior of an observer might be influenced by the positive or negative consequences of a behavioral model. An observer will emulate the behavior of a model, if this includes characteristics which the observer deems attracting or desirable, such as talent, intelligence, power etc. Furthermore, the way through which the model is treated will influence the observer. If the model is rewarded, it is more probable that the observer will emulate the rewarded behavior, while the opposite is expected to happen when an observed behavior is reprimanded. The brain activity when a subject performs an action or observes the actions of other people can be analyzed by means of Event Related Potentials (ERPs). ERPs are a special category of electroencephalographic (EEG) signals, which are recorded from various locations on a subject’s scalp when the subject is presented with external stimuli or events. ERPs provide non-invasive measurements of the electrical activity of the brain and describe the specific cognitive processes that are responsible for processing the stimuli or the events. The reliable detection of correct/incorrect actions is the basis for the implementation of brain computer interface (BCI) systems that decode brain electrical activity into actions controlling devices that will assist the users of the system. Identification of Error-Related Observations from Event Related Potentials Using Pattern Recognition Techniques This paper provides comparative results from the application of five pattern recognition techniques on the task of discriminating between observations of correct and incorrect actions using ERPs. Materials and Methods The ERP data used in the present study were collected in previous research (van Schie et al., 2004). The data were acquired from sixteen (16) healthy volunteers (observers), who observed correct or incorrect responses of subjects (actors) performing a special designed task. In particular, the actors were seated in front of a table facing an observer, having in front of them, on the table, two joystick devices positioned to the left and right of a LED stimulus device. The actors were asked to respond to the direction of a center arrowhead surrounded by distracting flankers pointing either in the same direction as the center arrow, or in opposite direction (Fig. 1).