Título: Randomized Pattern Classifiers for Epileptic Seizure Detection: A Performance Comparison
Autores: Silva, Natanael;Peixoto, Julio;Barreto, Guilherme
In this paper we evaluate the performances of randomized pattern classifiers in the task of EEG-based epileptic seizures detection. Our goal is to investigate if these new class of machine learning methods actually outperform powerful nonlinear classifiers, such as the MLP and SVM, in complex pattern recognition tasks. The rationale for the current work comes from the observation that the recent wave of applications involving randomized classifiers tend to report only positive reports, in which these networks always achieve equivalent or better performances than non-randomized nonlinear classifiers. A comprehensive performance evaluation is carried out, with the results strongly corroborate our hypothesis that randomized classifiers usually do not perform better than well trained standard nonlinear classifiers. Additionally, the performances of randomized classifiers are more dependent on the feature extraction method than non-randomized ones.
epileptic seizures;randomized classifiers;Welch’s periodogram;LPC coefficients;ROC curves
Código DOI: 10.21528/CBIC2017-57
Artigo em pdf: cbic-paper-57.pdf
Arquivo BibTeX: cbic-paper-57.bib