Classification of EEG Motion Artifact Signals Using Spatial ICA

Abstract

Using electroencephalography (EEG) data to extract information about cortical signals has become an increasingly explored task of interest in the field of computational neuroscience. In this paper, we proposed a novel procedure which reduce dimension by applying spatial Independent Component Analysis (SICA) on EEG motion artifact data and classify gait speed for a given subject by the projected EEG motion artifact signals. Whereas most applications of ICA in analyzing EEG data employ temporal ICA, we use SICA and Principal Component Analysis for dimension reduction before applying classifiers such as Support Vector Machines, Naive Bayes, and multinomial logistic regression to the extracted independent components. We evaluate and compare the classification models by using randomly selected channels from the multi-channel EEG motion artifact data as our test data. For practical application and interpretation, we treat the test channels as if they might come from a new trial for the given subject.