Fuzzy k-NN classification with weights modified by most informative neighbors of nearest neighbors

Nilgun Guler Bayazit*, Ulug Bayazit

*Bu çalışma için yazışmadan sorumlu yazar

Araştırma sonucu: ???type-name???Makalebilirkişi

4 Atıf (Scopus)

Özet

In the conventional fuzzy k-NN classification rule, the vote cast by each nearest neighboring known (labelled) sample on the class membership grades of the unknown (unlabelled) sample is formed by weighting the nearest neighbor's class membership grades by the inverse of the nearest neighbor's distance to the unknown sample. This paper proposes a modification of the weight (distance) used for each nearest neighbor by employing the geometrical relation among the nearest neighbor, its most informative known neighbor of the same class and the unknown sample. It is also proposed that this modification be only (conditionally) applied when the feature vector of the unknown sample lies outside the convex hull of the feature vectors of the known samples of each class. Results on a large number of datasets from the UCI and KEEL repositories and synthetically generated datasets show that, in return for a modest increase in classification complexity over the original fuzzy k-NN rule, the proposed fuzzy k-NN rule offers a better classification accuracy than the accuracies of the original fuzzy kNN rule and most other nearest neighbor type algorithms.

Orijinal dilİngilizce
Sayfa (başlangıç-bitiş)6717-6729
Sayfa sayısı13
DergiJournal of Intelligent and Fuzzy Systems
Hacim36
Basın numarası6
DOI'lar
Yayın durumuYayınlandı - 2019

Bibliyografik not

Publisher Copyright:
© 2019 IOS Press and the authors. All rights reserved.

Parmak izi

Fuzzy k-NN classification with weights modified by most informative neighbors of nearest neighbors' araştırma başlıklarına git. Birlikte benzersiz bir parmak izi oluştururlar.

Alıntı Yap