An efficient incremental mining algorithm for compact realization of prototypes

Viswanath, P and Murty, Narasimha M (2002) An efficient incremental mining algorithm for compact realization of prototypes. UNSPECIFIED.

Preview

Postscript viswanath-murty-csa-techreport.ps Download (148Kb)

Abstract

There are two phases in pattern classi viz design phase (abstractions are created/learnt), classi phase (abstractions are used to classify a test pattern). Classi based on neural networks and genetic algorithms needs more design time. Where as nearest neighbor classi have zero design time, they need more classi time. PC-tree (pattern count tree) based classifiers trikeskes a balance between the above mentioned two categories. PC-tree is a data structure originally used in incremental data mining to store the database. It can also be used in pattern classi for compact realization of prototypes. PC-Tree based classi has advantages over con- ventional classi nearest neighbor, k-nearest neighbor classi with respect to storage space, classi time, and classi accuracy(CA). In this paper we propose a classi algorithm which is superior to PC-tree based classi in every aspect. This classi is based on a new data structure called partitioned pattern count tree (PPC-tree). PPC-tree is more compact than PC-tree. This results in reduction of space required and classi time. The e number of training patterns stored in PPC -tree is in orders more than the actual number of training patterns, this results in increase of the CA.

Item Type:	Departmental Technical Report
Keywords:	PC Tree;PPC Tree
Department/Centre:	Division of Electrical Sciences > Computer Science & Automation (Formerly, School of Automation)
Date Deposited:	23 Jan 2007
Last Modified:	19 Sep 2010 04:12
URI:	http://eprints.iisc.ernet.in/id/eprint/15

Actions (login required)

View Item