Granular description of data: Building information granules with the aid of the principle of justifiable granularity
Granular description of data: Building information granules with the aid of the principle of justifiable granularity
复制标题
DOI：
10.1109/fuzz-ieee.2016.7737793
复制
发表时间：
2016-07
期刊：
影响因子：
--
通讯作者：
Xiubin Zhu;W. Pedrycz;Zhiwu Li
中科院分区：
文献类型：
--
作者：
Xiubin Zhu;W. Pedrycz;Zhiwu Li
作为反映领域知识和实验数值证据的通用构建块，信息颗粒在颗粒计算中实现的处理和促进与环境的通信中起着关键作用。在这项研究中，我们关注的是一个基本问题，即使用合理粒度原则构建一个有意义的、易于解释的球形信息颗粒集合。设计过程被表述为一个优化问题。首先，确定一系列数字原型，围绕这些原型构建信息颗粒。其次，以某一性能指标最大化为目标，对这些信息颗粒的半径值进行优化。比较了两种确定信息颗粒中心的方法，即随机选择的数值原型和借助聚类生成的原型。介绍并研究了两种优化准则。实验研究涉及合成数据以及来自UCI机器学习存储库的数据。
Formed as generic building blocks being reflective of domain knowledge and experimental numeric evidence, information granules play a pivotal role in processing realized in Granular Computing and facilitating communication with the environment. In this study, we are concerned with a fundamental problem of constructing a collection of meaningful, easily interpretable spherical information granules with the use of the principle of justifiable granularity. The design process is formulated as an optimization problem. First, a series of numeric prototypes are determined around which information granules are constructed. Second, the values of radii of these information granules are optimized aiming at maximizing a certain performance index. Two alternatives of determining centers of information granules are compared, i.e., randomly selected numeric prototypes and prototypes generated with the aid of clustering. Two optimization criteria are also introduced and studied. Experimental studies involving synthetic data as well as data coming from the UCI Machine Learning repository are reported.