基于相对分辨能力的属性约简算法

在粗糙集理论中,分辨能力反映拥有知识的多少;为此,给出分辨能力相关概念、性质和计算方法,并提出基于相对分辨能力的约简定义,同时研究该约简定义与Hu差别矩阵约简之间的等价性,指出Hu差别矩阵约简可由相对分辨能力约简获得.为了进一步提高求解效率,通过减少约简过程中基数排序次数来提升效率,设计了相对分辨能力的约简算法,其时间复杂度为O(|C|~2|U|).实例分析和UcI中数据集的实验比较表明所提出的约简算法是有效的、可行的.

Attribute reduction algorithm based on relative discernibility ability

In the rough set theory, discernibility ability represents that how much knowledge it holds. Firstly, some notions, important characters and computational methods corresponding to the discernibility ability are present. The definition of attribute reduction based on relative discernibility ability is proposed, and it is pointed that the reduction definition is consistent with the reduction of Hu's discernibility matrix. Therefore the reduction of Hu's discernibility matrix can be derived from the reduction of the relative discernibility ability. In order to further enhance the efficiency of attribute reduction, an effective attribute reduction algorithm based on relative discernibility ability is designed, which is improved through reducing the number of the radix sort, and the time complexity of the reduction algorithm is O(|C|²|U|). Finally, the example and experiments results from datasets of UCI demonstrate that the proposed algorithms are effective and feasible.