Extended k-Nearest Neighbours based on Evidence Theory

H Wang, DA Bell

Research output: Contribution to journalArticlepeer-review

24 Citations (Scopus)
Original languageEnglish
Pages (from-to)662-672
JournalThe Computer Journal
Issue number6
Publication statusPublished (in print/issue) - 26 Jun 2004

Bibliographical note

Other Details
Denoeux's evidence theoretic classifier needs to generate multiple mass functions, which are then combined and used for classification. The combination operation is known to be computationally expensive. A novel evidence theoretic classifier is proposed in this paper where only one mass function is needed to compute the class conditional probability for classification. Such a classifier extends the standard majority voting k-nearest neighbour classifier, and is an approximation of the Bayes classifier. This classifier is shown to perform better than the voting and distance weighted k-nearest neighbour classifiers. This work has led to a series of papers on neighbourhood counting methodology.

Cite this