A fast fuzzy $K$-nearest neighbour algorithm for pattern classification

  • Authors:

  • Yiannis S. Boutalis;Ioannis T. Andreadis;George D. Tambakis

  • Affiliations:

  • Section of Electronics and Information Systems Technology, Department of Electrical and Computer Engineering, Democritus University of Thrace, GR-67100, XANTHI, Greece. E-mail: ybout@demokritos.cc ...;Section of Electronics and Information Systems Technology, Department of Electrical and Computer Engineering, Democritus University of Thrace, GR-67100, XANTHI, Greece. E-mail: ybout@demokritos.cc ...;Section of Electronics and Information Systems Technology, Department of Electrical and Computer Engineering, Democritus University of Thrace, GR-67100, XANTHI, Greece. E-mail: ybout@demokritos.cc ...

  • Venue:
  • Intelligent Data Analysis
  • Year:
  • 2000

Quantified Score

Hi-index 0.01

Visualization

Abstract

A fast procedure for classifying a given test pattern to one ofits possible classes using both the K-NN decision rule and conceptsof the fuzzy set theory is described in this paper. The method isdivided into two steps; in the first step the K nearest neighboursare found using a fast procedure, whereas in the second step thetest pattern is classified using a fuzzy distance measure. The fastK-NN algorithm proceeds in finding a region containing at least Kneighbours around the test sample by utilising an ordered searchprocedure of the test point from three reference points. In thesequence the found region is modified in such a way that the real Knearest neighbours of the given point will be always inside it.Then a small number of distance calculations are required toidentify the true K-nearest neighbours and the fuzzy measure isapplied to classify the test pattern. The pre-processing load isquite moderate and computer simulation results show that themisclassification rate is lower than, or similar to, the crispversion, while presenting results richer in information contentthan its crisp counterpart. This rate is also kept low even in thecase we do not perform modifications that ensure the true K-NNfinding.