TY - JOUR
T1 - Rough set-based classification of EEG-signals to detect intraoperative awareness
T2 - 4th International Conference, RSCTC 2004
AU - Ningler, Michael
AU - Stockmanns, Gudrun
AU - Schneider, Gerhard
AU - Dressler, Oliver
AU - Kochs, Eberhard F.
PY - 2004
Y1 - 2004
N2 - Automated classification of calculated EEG parameters has been shown to be a promising method for detection of intraoperative awareness. In the present study, rough set-based methods were employed to generate classification rules. For these methods, discrete attributes are required. We compared a crisp and a fuzzy discretization of the real parameter values. Fuzzy discretization transforms one real attribute value to several discrete values. By combining the different (discrete) values of all attributes, several sub-objects were produced from a single original object. Rule generation from a training set of objects and classification of a test set provided good classification rates of approximately 90% for both crisp and fuzzy discretization. Fuzzy discretization resulted in a simpler and smaller rule set than crisp discretization. Therefore, the simplicity of the resulting classifier justifies the higher computational effort caused by fuzzy discretization.
AB - Automated classification of calculated EEG parameters has been shown to be a promising method for detection of intraoperative awareness. In the present study, rough set-based methods were employed to generate classification rules. For these methods, discrete attributes are required. We compared a crisp and a fuzzy discretization of the real parameter values. Fuzzy discretization transforms one real attribute value to several discrete values. By combining the different (discrete) values of all attributes, several sub-objects were produced from a single original object. Rule generation from a training set of objects and classification of a test set provided good classification rates of approximately 90% for both crisp and fuzzy discretization. Fuzzy discretization resulted in a simpler and smaller rule set than crisp discretization. Therefore, the simplicity of the resulting classifier justifies the higher computational effort caused by fuzzy discretization.
UR - http://www.scopus.com/inward/record.url?scp=9444229364&partnerID=8YFLogxK
U2 - 10.1007/978-3-540-25929-9_105
DO - 10.1007/978-3-540-25929-9_105
M3 - Conference article
AN - SCOPUS:9444229364
SN - 0302-9743
VL - 3066
SP - 825
EP - 834
JO - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
JF - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Y2 - 1 June 2004 through 5 June 2004
ER -