Abstract: A new approach to analysis of structure of the training sample based on identification and parameterization clusters of local regularities that are considered as generalized precedents of manifestation of partial interrelations in data is investigated. Substantive treatment of non-uniformity in images of empirical distributions in parametric spaces is proposed, and possibilities of use of secondary cluster structure for reduction of complexity of decisions and increase of processing speed, identification and verification of regularities, are studied.

Keywords: local dependency, generalized precedent, parametric space, cluster, hyper-parallelepiped, logical regularity, hypercube bitmap, derivative distribution, decision rule

Link:

ASSEMBLING DECISION RULE ON THE BASE OF GENERALIZED PRECEDENTS

Vladimir Ryazanov, Alexander Vinogradov, Yuryi Laptin

http://www.foibg.com/ijita/vol23/ijita23-03-p04.pdf