Title :
k nearest neighbors in search of a metric
Author :
Snapp, Robert R. ; Venkatesh, Santosh S.
Author_Institution :
Dept. of Comput. Sci. & Electr. Eng., Vermont Univ., Burlington, VT, USA
Abstract :
The finite-sample risk of the k-nearest neighbor classifier that uses a weighted Lp metric as a measure of class similarity is examined. For a family of multiclass, classification problems with smooth distributions in Rn, the risk is represented as an asymptotic expansion in decreasing fractional powers of the reference sample size. An analysis of the leading coefficients reveals that the optimal metric (i.e., the metric that minimizes the risk) tends to a weighted Euclidean (i.e., L2) metric as the sample size is increased. Numerical calculations corroborate this finding
Keywords :
pattern classification; random processes; asymptotic expansion; class similarity; finite-sample risk; k-nearest neighbor classifier; metric; multiclass classification problem; optimal metric; reference sample size; smooth distributions; weighted Euclidean metric; Computer science; Costs; Electric variables measurement; Laboratories; Nearest neighbor searches; Random processes; Risk analysis; Symmetric matrices; Testing;
Conference_Titel :
Information Theory, 1995. Proceedings., 1995 IEEE International Symposium on
Conference_Location :
Whistler, BC
Print_ISBN :
0-7803-2453-6
DOI :
10.1109/ISIT.1995.535771