Weighting of the k-Nearest-Neighbors

Konstantin Chernoff; Mads Nielsen

doi:10.1109/ICPR.2010.168

Weighting of the k-Nearest-Neighbors

Konstantin Chernoff, Mads Nielsen

Department of Computer Science

Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review

11 Citations (Scopus)

Abstract

This paper presents two distribution independent weighting schemes for k-Nearest-Neighbors (kNN). Applying the first scheme in a Leave-One-Out (LOO) setting corresponds to performing complete b-fold cross validation (b-CCV), while applying the second scheme corresponds to performing bootstrapping in the limit of infinite iterations. We demonstrate that the soft kNN errors obtained through b-CCV can be obtained by applying the weighted kNN in a LOO setting, and that the proposed weighting schemes can decrease the variance and improve the generalization of kNN in a CV setting.

Original language	English
Title of host publication	2010 20th International Conference on Pattern Recognition (ICPR)
Number of pages	4
Publisher	IEEE
Publication date	2010
Pages	666-669
ISBN (Print)	978-1-4244-7542-1
ISBN (Electronic)	978-1-4244-7541-4
DOIs	https://doi.org/10.1109/ICPR.2010.168
Publication status	Published - 2010
Event	20th International Conference on Pattern Recognition - Istanbul, Turkey Duration: 23 Aug 2010 → 26 Aug 2010

Conference

Conference	20th International Conference on Pattern Recognition
Country/Territory	Turkey
City	Istanbul
Period	23/08/2010 → 26/08/2010

Access to Document

10.1109/ICPR.2010.168

Cite this

@inproceedings{a617c540f16911dfb6d2000ea68e967b,

title = "Weighting of the k-Nearest-Neighbors",

abstract = "This paper presents two distribution independent weighting schemes for k-Nearest-Neighbors (kNN). Applying the first scheme in a Leave-One-Out (LOO) setting corresponds to performing complete b-fold cross validation (b-CCV), while applying the second scheme corresponds to performing bootstrapping in the limit of infinite iterations. We demonstrate that the soft kNN errors obtained through b-CCV can be obtained by applying the weighted kNN in a LOO setting, and that the proposed weighting schemes can decrease the variance and improve the generalization of kNN in a CV setting. ",

author = "Konstantin Chernoff and Mads Nielsen",

year = "2010",

doi = "10.1109/ICPR.2010.168",

language = "English",

isbn = "978-1-4244-7542-1",

pages = "666--669",

booktitle = "2010 20th International Conference on Pattern Recognition (ICPR)",

publisher = "IEEE",

note = "null ; Conference date: 23-08-2010 Through 26-08-2010",

}

TY - GEN

T1 - Weighting of the k-Nearest-Neighbors

AU - Chernoff, Konstantin

AU - Nielsen, Mads

PY - 2010

Y1 - 2010

N2 - This paper presents two distribution independent weighting schemes for k-Nearest-Neighbors (kNN). Applying the first scheme in a Leave-One-Out (LOO) setting corresponds to performing complete b-fold cross validation (b-CCV), while applying the second scheme corresponds to performing bootstrapping in the limit of infinite iterations. We demonstrate that the soft kNN errors obtained through b-CCV can be obtained by applying the weighted kNN in a LOO setting, and that the proposed weighting schemes can decrease the variance and improve the generalization of kNN in a CV setting.

AB - This paper presents two distribution independent weighting schemes for k-Nearest-Neighbors (kNN). Applying the first scheme in a Leave-One-Out (LOO) setting corresponds to performing complete b-fold cross validation (b-CCV), while applying the second scheme corresponds to performing bootstrapping in the limit of infinite iterations. We demonstrate that the soft kNN errors obtained through b-CCV can be obtained by applying the weighted kNN in a LOO setting, and that the proposed weighting schemes can decrease the variance and improve the generalization of kNN in a CV setting.

U2 - 10.1109/ICPR.2010.168

DO - 10.1109/ICPR.2010.168

M3 - Article in proceedings

SN - 978-1-4244-7542-1

SP - 666

EP - 669

BT - 2010 20th International Conference on Pattern Recognition (ICPR)

PB - IEEE

Y2 - 23 August 2010 through 26 August 2010

ER -