面向动态混合数据的多粒度增量特征选择算法  

Multi-granulation Incremental Feature Selection Algorithm for Dynamic Hybrid Data

作　　者：王锋[1] 姚珍 梁吉业[1] WANG Feng;YAO Zhen;LIANG Ji-Ye(School of Computer and Information Technology,Shanxi University,Taiyuan 030006,China)

机构地区：[1]山西大学计算机与信息技术学院,山西太原030006

出　　处：《软件学报》2025年第3期1186-1201,共16页Journal of Software

基　　金：国家自然科学基金(62276158);山西省回国留学人员科研项目(2021-007)。

摘　　要：在大数据时代,样本规模以及维数的动态更新和变化极大地增加了计算负担,在这些动态数据中,大多的数据样本并不以单一的数据取值形式存在,而是同时包含符号型数据和数值型数据的混合型数据.为此,学者们提出了许多关于混合数据的特征选择算法,但现有的算法大多只适用静态数据或者小规模的增量数据,无法处理大规模动态变化的数据,尤其是数据分布不断变化的大规模增量数据集.针对这一局限性,通过分析动态数据中粒空间以及粒结构的变化和更新,基于信息融合机制,提出了一种面向动态混合数据的多粒度增量特征选择算法.该算法重点讨论了动态混合数据中的粒空间构建机制、多数据粒结构的动态更新机制以及面向数据分布变化信息融合机制.最后,通过与其他算法在UCI数据集上的实验结果进行对比,进一步验证了所提算法的可行性和高效性.In the era of big data,the sample scale and the dynamic update and variation of dimensionality greatly increase the computational burden.Most of these data sets do not exist in the form of a single data type but are more often hybrid data containing both symbolic and numerical data.For this reason,scholars have proposed many feature selection algorithms for hybrid data.However,most of the existing algorithms are only applicable to static data or small-scale incremental data and cannot handle large-scale dynamic changing data,especially large-scale incremental data sets with changing data distribution.To address this limitation,this paper proposes a multi-granulation incremental feature selection algorithm for dynamic hybrid data based on an information fusion mechanism by analyzing the variations and updates of granularity space and granularity structure in dynamic data.The algorithm focuses on the mechanism of granularity space construction in dynamic hybrid data,the mechanism of dynamic update of multiple data granularity structures,and the mechanism of information fusion for data distribution variations.Finally,the paper verifies the feasibility and efficiency of the proposed algorithm by comparing the experimental results with other algorithms on the UCI dataset.

关 键 词：动态混合数据 数据分布变化 多粒度计算 信息融合 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]