Multi-view Learning for EEG Signal Classification of Imagined Speech

Sandra Eugenia Barajas Montiel, Eduardo F. Morales, Hugo Jair Escalante

Abstract: Multi-view Learning (MVL) has the objective of combining the information that describes an object from different groups of features. This machine learning paradigm has proven useful to improve generalization performance of classifiers by taking advantage of the complementary information from different views of the same object. This work explores the use of three Co-training-based methods and three Co-regularization techniques to perform supervised learning to classify electroencephalography signals (EEG) of imagined speech. Two different views were used to characterize these signals, extracting Hjorth parameters and the average power of the signal. The results of six different approaches of MVL applied to classify the imagined speech of five different words are reported, showing an improvement up to 14.27% in accuracy average of classification compared with single view classification.