A novel method for single-trial classification in the face of temporal variability

Abstract

Machine learning techniques have been used to classify patterns of neural data obtained from electroencephalography (EEG) to increase human-system performance. This classification approach works well in controlled laboratory settings since many of the machine learning techniques used often rely on consistent neural responses and behavioral performance over time. Moving to more dynamic, unconstrained environments, however, introduces temporal variability in the neural response resulting in sub-optimal classification performance. This study describes a novel classification method that accounts for temporal variability in the neural response to increase classification performance. Specifically, using sliding windows in hierarchical discriminant component analysis (HDCA), we demonstrate a decrease in classification error by over 50% when compared to other state-of-the-art classification methods. © 2013 Springer-Verlag Berlin Heidelberg.