Deleting or keeping outliers for classifier training?

Author

Tallon-Ballesteros, Antonio J. ; Riquelme, Jose C.

Author_Institution

Dept. of Languages & Comput. Syst., Univ. of Seville, Seville, Spain

fYear

2014

fDate

July 30 2014-Aug. 1 2014

Firstpage

281

Lastpage

286

Abstract

This paper introduces two statistical outlier detection approaches by classes. Experiments on binary and multi-class classification problems reveal that the partial removal of outliers improves significantly one or two performance measures for C4.5 and 1-nearest neighbour classifiers. Also, a taxonomy of problems according to the amount of outliers is proposed.

Keywords

data mining; statistical analysis; binary classification problems; classifier training; data mining; multiclass classification problems; nearest neighbour classifiers; statistical outlier detection; Detectors; Training; attribute noise; classification; inter-quartile range; outlier detection; partial removal; statistical outlier detection;

fLanguage

English

Publisher

ieee

Conference_Titel

Nature and Biologically Inspired Computing (NaBIC), 2014 Sixth World Congress on

Conference_Location

Porto

Print_ISBN

978-1-4799-5936-5

Type

conf

DOI

10.1109/NaBIC.2014.6921892

Filename

6921892

Link To Document

https://search.isc.ac/dl/search/defaultta.aspx?DTC=49&DC=117271