基于改进差别信息树的广义决策属性约简

Generalized Decision Attribute Reduction Based on Improved Discernibility Information Tree

属性约简作为一种有效的数据降维方法,对于处理高维数据具有重要意义,通过删除冗余属性保留重要属性,获得与原系统具有相同表达能力和分类能力的属性子集。差别矩阵是得到属性约简的一种重要方法,但其中含有大量无用的信息,本文受改进差别信息树的启发,将改进差别信息树与决策多层次系统相结合,在该方法下研究不同决策层级间改进差别信息树之间的关系,提出一种基于改进差别信息树的广义决策属性约简算法。所提方法不仅可以实现对差别矩阵中非空元素的压缩存储,还有效缩短了时间消耗。为了验证算法的有效性,选取8组UCI数据集分别从算法的约简结果和约简效率两方面进行对比,实验结果验证了算法的可行性和有效性。

Attribute reduction, as an effective data dimensionality reduction method, is of great significance for dealing with high-dimensional data, which obtains a subset of attributes with the same expressive and categorization ability as the original system by removing redundant attributes and retain-ing important attributes. Discernibility matrix is an important method to get attribute reduction, but it contains a lot of useless information, this paper is inspired by the improved discernibility information tree, combines the improved discernibility information tree with the multi-hierarchical decision systems, studies the relationship between the improved discernibility information tree among different decision levels under this method, and proposes a generalized decision attribute reduction algorithm based on the improved discernibility information tree. The proposed method can not only realize the compressed storage of non-empty elements in the discernibility matrix, but also effectively reduce the time consumption. In order to verify the effectiveness of the algorithm, eight groups of UCI datasets are selected to compare the algorithm in terms of reduction results and reduction efficiency, and the experimental results verify the feasibility and effectiveness of the algorithm.