区间值决策系统下最短约简算法的研究

Research on Shortest Reduction Algorithm in Interval-Valued Systems

摘要: 属性约简可以选出保持分类能力不变的属性子集，而最短约简不仅可以选出保持分类能力不变的属性子集，还可以最大程度地删除冗余属性、压缩决策表，选出最优的属性子集。本文在区间值决策系统的数据背景下，分别对针对决策属性的全部决策类和特定决策类构建二进制差别矩阵，结合SRA算法分别提出了基于二进制差别矩阵的最短约简算法和特定类最短约简算法。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长度和约简效率两方面进行对比，实验结果证明了算法的可行性和有效性。

Abstract: Attribute reduction can select a subset of attributes that maintains the classification ability, while shortest reduction can not only select a subset of attributes that maintains the classifica-tion ability, but also delete redundant attributes and compress decision tables to select the optimal subset of attributes. In this paper, based on the data background of interval-valued decision systems, binary discernibility matrix were constructed for all decision classes and specific decision classes, and the shortest reduction algorithm and the specific class shortest reduction algorithm based on binary discernibility matrix were proposed, respectively, combined with the SRA algorithm. To verify the effectiveness of the algorithm, 8 UCI datasets were selected for comparison from the perspective of the length of the reduction result and the efficiency of the reduction algorithm. The experimental results prove the feasibility and effectiveness of the al-gorithm.