计算机工程

• •    

结合区间二型FRCM与混合度量的两阶段信息粒化

  

  • 发布日期:2020-12-15

Two-phase Information Granulation combined with Interval Type-2 FRCM and Mixed Metrics

  • Published:2020-12-15

摘要: 作为粒计算的基础,信息粒化受到越来越多的关注。针对类簇交叉且分布不均衡数据的信息粒化问题,在可信粒度 准则的两阶段信息粒化框架下,第一阶段基于区间二型模糊粗糙 C 均值算法对交叉类簇分布不均衡数据进行聚类分析,得到 初始的信息粒;第二阶段综合考虑数据空间分布、样本规模及粒子性质等因素,采用混合度量的方法,设计均衡证据合理性 和语义独特性的粒化函数,并基于参数版可信粒度准则优化由覆盖度和独特性组成的复合函数,求解最佳粒子边界。人工数 据集和不同 UCI 数据集的仿真实验结果表明,所提算法有效提高了不平衡数据的信息粒化质量及粒子代表性,在归类正确数、 粒子特性等指标上均取得了理想表现。

Abstract: As the basis of granular computing, information granularity has received more and more attention. To solve the information granularity of the crossed data with unbalanced distribution, the two-phase framework of information granularity based on the principle of justifiable granularity is introduced. For the crossed and unbalanced data, in the first phase, an interval type-2 fuzzy rough C-means algorithm is given for clustering analysis and getting initial information granules. In the second phase, considering the spatial distribution, sample size and characters of granules, granulation function with reasonable experimental evidence and well-defined semantics is designed based on the method of mixed metrics. Based on the principle of the parametric justifiable granularity, the best boundary of granule is obtained by optimizing the composite function combined by the coverage and specificity. The simulation results with the artificial datasets and different UCI datasets showed that the proposed algorithm achieves ideal result in classification accuracy and particle characteristics, it can solve the granulation of the unbalanced data and improve the quality of the granule and the representation of the granule.