Information Technology Reference
In-Depth Information
discover the associations and relations of internal data. However, although rough
set theory plays a good role on the processing of fuzzy and incomplete
knowledge, it has poor ability to deal with original fuzzy data. Therefore, the
integration between rough set theory and others, such as fuzzy set theory, neural
network theory and etc, will benefit its applications.
Rough set based KDD systems are usually composed by some parts, such as
data preprocessing, rough set or other extended theory based data reduction,
decision algorithms and etc. The general idea is that: first preprocess data and
prepare for data reduction, and then calculate reducts or approximate reducts; at
last, extract rules by value reduction algorithm (reduce the numbers of attributes
and objects) and apply the rules to new objects.
In the past few years, a lot of rough set based KDD systems are developed.
The most representative ones are LERS, ROSE, KDD-R, Rough Enough and etc.
1. LERS
LERS(Learning from examples based on Rough Set) system was developed by
Kansas university of USA, which is a rough set based case learning system
(Grzyrnala-busse, 1997). It is implemented on VAX9000 with Common Lisp.
LERS has been used in Johnson space center as an experts systems developing
tool for two years. Most of the developed experts systems can be used to the
iatric decision of space station. Moreover, LERS was also applied to the research
on environment protection, weather predictions and medical treatments.
2. ROSE
Poznan science and technology university of Poland developed ROSE (Rough
Set Data Explorer) system to have decision analysis. It is the new version of
Rough Das & Rough Class system, where Rough Das is to analyze information
of system data and Rough Class is to support the classification of objects. Rough
Das and Rough Class have been applied to many application fields. ROSE is an
interactive software system implemented on PC compatible machine and
Windows/NT system. The calculation modules of ROSE have the following
features: Data verification and preprocessing
Automatic discretization of continuous values with Fayyad and Irani
discretization algorithms
Qualitative estimation of condition attributes with traditional rough set models
or variable precision rough set models
Attribute core calculation and information table reduction with thealgorithms
developed by Romanski, Skowron and etc.
