1 INTRODUCTION
Attribute reduction, firstly proposed in rough set theory, is a
process to find the optimal subset of attributes that retains
the same discriminatory power of the whole attribute set
[8]. By eliminating the attributes that are unimportant or
irrelevant to the target concept, attribute reduction is often
carried out as a preprocessing step to reduce the storing
requirements and computational requirements by many
knowledge abstraction technologies, such as big
data[4-5,7], pattern recognition[1], and decision
analysis[9], etc.
As one of the most important contributions and challenges
in rough set theory, most attribute reduction methods are
dependent on discernibility or indiscernibility relation
([6,10,20]) and some extension classification conceptions,
such as positive region, boundary region and negative
region, etc. With the development of the research on
classification technology, a novel theory, called as
three-way decisions (3WD), is proposed in the last decade
and obtained the rapid development both in theory and
applications [2,19].
As the theory originated from rough set, 3WD has a tight
relationship to rough set theory. Many notions are shared by
them. For example, some 3WD models, such as the
classical 3WD model and the Pawlak three-way decisions
model [12], directly apply the rough set three-regions
(positive region, boundary region and negative region) to
divide a universe. It is noted that these shared notions are
This work is supported by National Nature Science Foundation under
Grant No.61502538, No.61290325, No.61773406; the Foundation for
Innovative Research Groups of the National Natural Science Foundation
of China (Grant No. 61621062); the Nature Science Foundation for Young
Scientists of Hunan Province, China (No.2015JJ3157).
also the basis of attribute reduction. This fact shows that
attribute reduction based on 3WD is also possible and
meaningful.
There are several 3WD models for different applications,
such as the classical 3WD model, evaluation-based 3WD
model, conceptual model of 3WD, and model of 3WD for
decision tables (3WD-D). Generally speaking, the
conceptual model is an abstract model. The evaluation-
based 3WD model is a generic model and depends on an
ordered set and a related evaluation function. In context of
classification, the classical 3WD model is only suitable for
information table, while 3WD-D is effective in decision
tables [16].
In this paper, we propose an attribute reduction method
based on 3WD-D, and compare with the traditional
attribute reduction methods based on rough set.
The remainder of this paper is structured as follows. Some
basic concepts are briefly reviewed in Section 2, which
include rough set, 3WD and 3WD-D. In Section 3, we
analyze the discernibility relations in 3WD-D and induce
three related discernibility matrices. On the basis, a
heuristic attribute reduction method is proposed in Section
4. Section 5 presents some experiments to validate the
efficiency of the proposed method and the differences from
the traditional methods based on rough set. Finally, we
conclude this paper and discuss the outlook for further work
in Section 6.
2 PRELIMINARIES
As a theory motivated by rough set three-regions, 3WD has
a tight relationship to rough set theory. In this section, we
briefly review the basic concepts and principles of rough set
in [14,17-18], 3WD in [3,11-13] and 3WD-D model in [16].
An Attribute Reduction Method Based On Three-Way Decisions Model for
Decision Tables
Linzi Yin
1
, Xuemei Xu
1
, Jiafeng Ding
1
,
Zhaohui Jiang
2
, Kehui Sun
1
1. School of Physics and Electronics, Central South University, Changsha 410083
E-mail: yinlinzi@csu.edu.cn, xuxuemei999@126.com,
csjfding@126.com, kehui@csu.edu.cn
2. School of Information Science and engineering, Central South University, Changsha 410083, China
E-mail: jzh0903@csu.edu.cn
Abstract: In this paper, an attribute reduction method is proposed based on the model of three-way decisions for decision
tables (3WD-D). First, the discernibility relations among regions are analyzed based on the component of trisecting in
3WD-D model, and the discernibility relations among the granules in the same regions are discussed based on the other
component of acting. On the basis, three discernibility matrices are presented to describe these discernibility relations
and calculate the significances of attributes. Next, a heuristic attribute reduction algorithm based on discernibility matrix
is proposed for finding a single reduct. Besides, the differences and relationships between 3WD-D and rough set are
analyzed in terms of discernibility matrices, reducts and types of rule set. These theoretical analysis and experimental
results show the effectiveness and advantages of the proposed attribute reduction method.
Key Words: Three-Way Decisions, Attribute Reduction, 3WD-D, Discernibility Matrix
381
978-1-5386-1243-9/18/$31.00
c
2018 IEEE