Performing Feature Selection Before Removing Outliers To Increase Classfier's Accuracy

This publication doesn't include Faculty of Arts. It includes Faculty of Informatics. Official publication website can be found on muni.cz.

Authors

PETLIAK Nataliia TMENOVA Oleksandra NÁMEŠNÝ Matúš BONČO Tomáš POPELÍNSKÝ Lubomír

Type Article in Proceedings
Conference DATA A ZNALOSTI & WIKT 2018, sborník konference
MU Faculty or unit

Faculty of Informatics

Citation
Keywords Feature selection; Outlier detection; classification accuracy
Description This work addresses the problem of feature selection for boosting the performance of outlier detectors in the context of supervised classification. Different feature selection and outlier detection methods are applied to four datasets used in the experiment and a comparative analysis between combinations of these methods is reported. We present combinations producing the best accuracy of a classifier and show the optimal number of outliers to be removed.
Related projects: