Abstract: There are different applications that require finding in a set all points most similar to the given query element. These are problems known as classification problems, or as best match, or the nearest neighbor search procedures. The main goal of this paper is to analyze the algorithmic constructions that appear in a model of recognizing the nearest neighbors in binary data sets. Our target is an earlier proposed algorithm W,1971, R 1974, A,2000 where existence and effectiveness of structures that serve the search procedure is raised. These structures split the search area into the perfect codes and/or the discrete isoperimetry problem solutions, which coexist for a very limited area of parameters. We extend indirectly these parameters through the composite block constructions and estimate the resulting loss of effectiveness of the procedure under consideration.

Keywords: Pattern Recognition, Classification, Best Match, Perfect Code

ACM Classification Keywords: 1.5. Pattern recognition, H.2.8 Database applications, Data mining

Link:

COMPOSITE BLOCK OPTIMIZED CLASSIFICATION DATA STRUCTURES

Levon Aslanyan, Hasmik Sahakyan

http://foibg.com/ibs_isc/ibs-16/ibs-16-p03.pdf