Title

A Novel Multivariate Analysis Method to Classify Different Tasks Based on Rapid Electroencephalography (Eeg) Data

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer Science and Info Sys

Date of Award

Summer 2015

Abstract

Multi-channel electroencephalography (EEG) data are temporal electrical signals which provide information about the neural activity in different parts of brain cortex. Analyses of multi-channel spatio-temporal EEG data which are collected while the brain is engaged with certain tasks or stimuli can provide information about neural signatures associated with those tasks or stimuli. In particular, machine learning / pattern classification algorithms can be applied to classify different classes of tasks or stimuli that the brain is engaged with. There are many machine learning algorithms designed for analyzing spatio-temporal data, and almost all of them treat space and time dimensions of data separately. Many emerging applications of machine learning algorithms require learning in the form of input-output mappings where the inputs are the observed data in a specific time period, and the outputs are the classes associated with the specific time-period.In this thesis, development of a new classification method by extending the spatio-temporal data in the feature space using the temporally-augmented version of features is introduced. The inherent nature of the developed method lends itself to work better on spatio-temporal data with fine temporal resolution, such as the EEG data. For this purpose, after IRB approval, rapid EEG data from 12 participants (6 females and 6 males) were collected while they were observing positive and negative images in a random order, and the newly developed classification method was applied to classify whether the participants were observing positive or negative images. The results of the new classification method were compared with the normal classification methods. The augmented method provided slightly better results (4-6% more) than non-augmented version: overall average classification accuracy of up to 58%, female average classification of up to 63%, male average classification of up to 58%, and individual maximum classification accuracy of up to 87% were achieved.

Advisor

Unal Sakoglu

Subject Categories

Computer Sciences | Physical Sciences and Mathematics

Share

COinS