基于随机抽样的变精度邻域粗糙集特征选择  

Variable Precision Neighborhood Rough Set Feature Selection Based on Random Sampling

作　　者：沈林[1] SHEN Lin(Putian University,Putian 351100,China)

机构地区：[1]莆田学院

出　　处：《廊坊师范学院学报（自然科学版）》2019年第2期14-17,共4页Journal of Langfang Normal University(Natural Science Edition)

基　　金：2015年度福建省教育厅中青年项目“变精度邻域粗糙集的分类与约简及其应用研究”(JA15458)

摘　　要：在变精度邻域粗糙集的特征选择中,改进后的区分矩阵相较于依赖度,具有时间复杂度较低的优点,但由于空间占用高,限制了其在大规模数据上的应用。减小数据规模虽然可以大幅降低空间消耗,但存在信息量丢失的风险,为此,通过随机抽取多个小规模数据子集以降低空间占用,在分别进行特征选择后,选择权重高的特征子集进行了测试。实验结果证明,可以将空间占用降低2-3个数量级,并保持精度和抽样前相当。In feature selection of variable precision neighborhood rough sets, the improved identification matrix has a lower time complexity than dependency. However, due to the high occupancy of space, its application in large-scale data is limited. Reducing data size can greatly reduce space consumption, but there is a risk of information loss. In this paper, several small-scale data subsets are randomly extracted to reduce the space consumption, and then after feature selection of each subset, the high-weight attribute subsets are selected to test. The experimental results show that the space occupancy can be reduced by 2-3 orders of magnitude, and the accuracy is comparable to that before sampling.

关 键 词：邻域粗糙集 随机抽样 特征选择 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]