Pattern recognition method based on Rvachev functions with engineering applications
Abstract
A novel method of pattern classification based on Rvachev functions (or R-functions) is developed. The method is termed the R-cloud classifier. The R-cloud classification is a non-parametric approach that makes no assumptions about the data model and does not require class density estimates. The implementation utilizes the methodological apparatus of R-functions and allows for implicit analytical representation of decision surfaces. This makes the R-cloud classifier attractive from an application viewpoint. The separating surfaces are represented by R-clouds, which are built on the notions of the separating primitive and the separating bundle. The representation of R-clouds can be reduced in order to achieve better computational characteristics. The reduced representation of an R-cloud yields a set of key vectors, which contains the most important information about a given data set from class separability standpoint. The method was tested and validated with data from the problem of the detection of abnormal software processes in computer systems. The versatility of the proposed R-function method and preliminary simulation results suggest that the method has potential for further development and adaptation to particular problems.
Degree
Ph.D.
Advisors
Tsoukalas, Purdue University.
Subject Area
Computer science
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.