| 面向一致性样本的属性约简 |
| |
| 引用本文： | 高媛,陈向坚,王平心,杨习贝.面向一致性样本的属性约简[J].智能系统学报,2019,14(6):1170-1178. |
| |
| 作者姓名： | 高媛 陈向坚 王平心 杨习贝 |
| |
| 作者单位： | 1. 江苏科技大学 计算机学院, 江苏 镇江 212003;2. 江苏科技大学 理学院, 江苏 镇江 212003 |
| |
| 摘 要： | 作为粗糙集理论的一个核心内容，属性约简致力于根据给定的约束条件删除数据中的冗余属性。基于贪心策略的启发式算法是求解约简的一种有效手段，这一手段通常使用数据中的全部样本来度量属性的重要度从而进一步得到约简子集。但实际上，不同样本对于属性重要度计算的贡献是不同的，有些样本对重要度贡献不高甚至几乎没有贡献，且当数据中的样本数过大时，利用全部样本进行约简求解会使得时间消耗过大而难以接受。为了解决这一问题，提出了一种基于一致性样本的属性约简策略。具体算法大致由3个步骤组成，首先，将满足一致性原则的样本挑选出来；其次，将这些选中的样本组成新的决策系统；最后，利用启发式框架在新的决策系统中求解约简。实验结果表明:与基于聚类采样的属性约简算法相比，所提方法能够提供更高的分类精度。
|
| 关 键 词： | 属性约简 分类精度 聚类 一致性样本 集成 启发式算法 邻域粗糙集 多准则 |
Attribute reduction over consistent samples |
| |
| GAO Yuan,CHEN Xiangjian,WANG Pingxin,YANG Xibei.Attribute reduction over consistent samples[J].CAAL Transactions on Intelligent Systems,2019,14(6):1170-1178. |
| |
| Authors: | GAO Yuan CHEN Xiangjian WANG Pingxin YANG Xibei |
| |
| Affiliation: | 1. School of Computer, Jiangsu University of Science and Technology, Zhenjiang 212003, China;2. School of Science, Jiangsu University of Science and Technology, Zhenjiang 212003, China |
| |
| Abstract: | As one of the key topics in rough sets theory, attribute reduction aims to remove redundant attributes in a data set according to a given constraint condition. Based on greedy strategy, the heuristic algorithm is an effective strategy in finding reductions. Traditional heuristic algorithms usually need to scan all samples in a data set to compute the significance of attributes to further obtain a reduction. However, different samples have different contributions to the process of computing significance. Some samples have little relation to the significance, and some even have no contribution to the significance. Therefore, scanning all samples to compute reductions may require too much time, and the time may be unacceptable if the number of samples is too large. To fill such a gap, we have proposed an attribute reduction algorithm with sample selection, which is based on the consistent principle. The algorithm is composed of three stages. First, the samples that satisfy the consistent principle were selected; second, a new decision system was constructed with these selected samples; finally, reductions were derived from the heuristic algorithm over the new decision system. Experimental results demonstrated that, compared with the attribute reduction algorithm with a cluster-based sample selection, our new algorithm can offer better classification accuracy. |
| |
| Keywords: | attribute reduction classification accuracy clustering consistent samples ensemble heuristic algorithm neighborhood rough set multiple criteria |
|
| 点击此处可从《智能系统学报》浏览原始摘要信息 |
|
点击此处可从《智能系统学报》下载全文 |
|
