TY  - RPRT
ID  - doell2013
T1  - Analysis of Electroencephalographic DWT Features for Classification and Regression of Visual Field Charts
A1  - Doell, Christoph
Y1  - 2013
T2  - Faculty of Computer Science, University of Magdeburg
AD  - Universit\"atsplatz 2, 39106 Magdeburg, Germany
N2  - This thesis is about with the analysis of correlations between brain activities of visually impaired subjects, expressed electroencephalographies (EEGs) and clinical variables relating to the visual perception, especially with the visual field of patients. While recent research on this topic interpret EEGs as brain graphs based on the synchronization likelihood, the primary goal of this work is to create a comparison to this approach using classical time series analysis on the EEGs. Secondary, we examine emerging questions by giving a detailed view on the data analysis process. We apply model-based
machine learning techniques and cross validation to measure the prediction quality of our generated models. We utilize discrete wavelet transformation (DWT ) and study different combinations of mother wavelets for feature generation. Within the presented approach, various regression algorithms including ordinary least squares, regression trees and support vector regression are compared. Further, two filtering concepts are investigated; the first depends on the noise rate of the EEG, the second on clinical
variables, limiting the reliability of the underlying tests of the visual field. We show that clinical variables are well predictable by using our features and a support vector regression model.
ER  -