Utilization of Singularity Exponent in Nearest Neighbor Based Classifier

0375796 - ÚI 2013 RIV US eng J - Journal Article
Jiřina, Marcel - Jiřina jr., M.
Utilization of Singularity Exponent in Nearest Neighbor Based Classifier.
Journal of Classification. Roč. 30, č. 1 (2013), s. 3-29. ISSN 0176-4268. E-ISSN 1432-1343
Grant - others:Czech Technical University(CZ) CZ68407700
Institutional support: RVO:67985807
Keywords : multivariate data * probability density estimation * classification * probability distribution mapping function * probability density mapping function * power approximation
Subject RIV: BB - Applied Statistics, Operational Research
Impact factor: 0.571, year: 2013

Classifiers serve as tools for classifying data into classes. They directly or indirectly take a distribution of data points around a given query point into account. To express the distribution of points from the viewpoint of distances from a given point, a probability distribution mapping function is introduced here. The approximation of this function in a form of a suitable power of the distance is presented. How to state this power - the distribution mapping exponent -- is described. This exponent is used for probability density estimation in high-dimensional spaces and for classification. A close relation of the exponent to a singularity exponent is discussed. It is also shown that this classifier exhibits significantly better behavior (classification accuracy) than other kinds of classifiers for some tasks.
Permanent Link: http://hdl.handle.net/11104/0208364