Data-guided multi-granularity selector for attribute reduction
- PDF / 1,667,434 Bytes
- 13 Pages / 595.224 x 790.955 pts Page_size
- 82 Downloads / 226 Views
Data-guided multi-granularity selector for attribute reduction Zehua Jiang1,2 · Huili Dou1,2 · Jingjing Song1,2 · Pingxin Wang3 · Xibei Yang1,2,4 · Yuhua Qian5
© Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Presently, the greedy searching strategy has been widely accepted for obtaining reduct in the field of rough set. In the framework of greedy searching, the evaluation of the candidate attribute is crucial, because the evaluation can determine the final result of reduct to a large extent. However, most of the previous evaluations are designed by considering one and only one fixed granularity, which fails to make the multi-view based evaluation possible. To fill such gap, a Parameterized Multigranularity Attribute Selector is proposed for obtaining reduct in this paper. Our attribute selector consists of two parts: one is the multi-granularity attribute selector which evaluates and selects attributes through using the information provided by multiple different granularities; the other is the data-guided parameterized granularity selector which generates multiple different parameterized granularities through taking the characteristics of data into account. The experimental results over 15 UCI data sets show the following: 1) compared with the state of the art approaches for obtaining reducts, our proposed attribute selector can contribute to reduct with higher stability; 2) our proposed attribute selector will not provide the reduct with poorer classification performance. This research suggests a new trend for the multi-granularity mechanism in the problem of attribute reduction. Keywords Attribute reduction · Data-guided · Parameterized multi-granularity
1 Introduction With the dramatically increasing of the dimensionality of data, much attention has been paid to the topic of attribute reduction [4, 5, 12, 23] in the field of the rough set theory [35, 36, 38, 42]. This can be attributed to the following facts: 1) attribute reduction can contribute to achieving an attribute subset (i.e., reduct) which does not contain irrelevant and redundant attributes; 2) such derived attribute subset may be equipped with some required semantic explanations because of the intended constraints defined in attribute reduction [57]. In recent years, many researchers have devoted to studying the approach of deriving reduct. With a careful reviewing of previous researches, from the viewpoint of Granular Computing [6, 16, 29, 33, 34, 46, 47], it is not difficult to conclude a similar structure among those approaches: given a granularity related constraint [10, 28], evaluate each candidate attribute over such granularity and then select a qualified
Huili Dou
[email protected]
Extended author information available on the last page of the article.
attribute. This process will be constantly executed until the given constraint is satisfied. For instance, if the neighborhood rough set [11, 21, 22, 50] is employed to generate reduct, then the constraint defined in attribute reduction is related to the gra
Data Loading...
