Rough classification and accuracy assessment

Abstract

In search for methods to handle imprecision in geographical information this paper explores the use of rough classification to represent uncertainty. Rough classification is based on rough set theory, where an uncertain set is specified by giving an upper and a lower approximation. Novel measures are presented to assess a single rough classification, to compare a rough classification to a crisp one and to compare two rough classifications. An extension to the error matrix paradigm is also presented, both for the rough-crisp and the roughrough cases. An experiment on vegetation and soil data demonstrates the viability of rough classification, comparing two incompatible vegetation classifications covering the same area. The potential uses of rough sets and rough classification are discussed and it is suggested that this approach should be further investigated as it can be used in a range of applications within geographic information science from data acquisition and analysis to metadata organization.