Incremental learning with hierarchies of rough decision tables

Abstract

Rough decision tables were introduced by Pawlak in the context of rough set theory. A rough decision table represents, a non-functional in general, relationship between two groups of properties of objects, referred to as condition and decision attributes, respectively. In practical applications, the rough decision tables are normally learned from data. In this process, for better coverage of the domain of interest, they can be structured into hierarchies. To achieve convergence of the learned hierarchy of the rough decision table to a stable final state, it is desirable to avoid total regeneration of the learned structure after new objects, not represented in the hierarchy, are encountered. This can be accomplished through an incremental learning process in which the structure of rough decision table's is updated, rather than regenerated, after new objects occurred. The presentation of this incremental learning process within the framework of the rough set model is the main theme of the article.