Exploring the most significant features for EEG ErrP detection through statistical analysis

Recently, electroencephalographic (EEG) signals have been used to design and enhance human-robot interaction (HRI). In particular, error-related potentials (ErrPs) have been leveraged since very few years. These potentials can be used to provide feedback to the robot about any mismatch between the user's expectations and the robot's behavior, during interaction tasks. In this process, the correct classification of ErrPs is crucial, which, in turn, relies on the reliability of the process for extraction and selection of signal features. In this work, we consider an extensive list of possible features and perform a statistical analysis to assess their discriminative power for ErrP analysis. The aim is to reduce the number of features used for classification while retaining the most relevant ones only. Overall, the outcome of our study shows that some parameters have relevant importance compared to others, (i.e. temporal features, frequency features, signal processing features, and some wavelet transform coefficients), while some of the features used in existing works are not useful since they have low discriminative power.