基于粒化可拓决策的属性约简算法研究

Research on Attribute Reduction Algorithm Based on Granulation Extension Decision

针对传统属性约简算法利用等价关系计算过程烦琐、运行时间较长的问题,定义能体现属性间相关程度的绝对关联度,提出一种基于粒化可拓决策的属性约简算法。首先,利用K-means聚类算法,对原始数据集进行粒化,得到各簇中心;其次,运用可拓决策理论确定经典域、节域和待评物元,通过计算各簇中属性之间的关联度构建指示矩阵,并计算各属性的指示值;最后,根据指示值,从大到小依次选择属性,实现样本集属性约简。实验结果表明,算法运算速度较快,约简后对数据集分类精度影响小,部分数据集分类精度有所提升,验证了算法的有效性。

Aiming to deal with the cumbersome calculation process and long running time of traditional attribute reduction algorithms using equivalence relations,an absolute correlation degree that could reflect the degree of correlation between attributes was defined,and an attribute reduction algorithm based on granulation extension decision was proposed.Firstly,K-means clustering algorithm was used to granulate the original data set to obtain the centers of each cluster.Secondly,the extension decision theory was used to determine the classical domain,the node domain and the matter to be evaluated.The indicator matrix was constructed by calculating the correlation degree between the attributes in each cluster,and the indicator value of each attribute was calculated.Finally,according to the indicated value,the attributes were selected in order from the largest to the smallest to realize the attribute reduction of the sample set.The experimental results showed that the algorithm had a faster operation speed.The reduction had small impact on the classification accuracy of the data set,while the classification accuracy of some data set had been improved,which verified the effectiveness of the algorithm.