基于改进辨识矩阵的变精度邻域粗糙集属性约简

提出一种用于变精度邻域粗糙集,可以大幅减少时间复杂度的属性约简算法.该算法基于一种改进的辨识矩阵.首先用辨识矩阵同时记录决策一致和不一致的数据,然后用二进制位运算计算样本的邻域,最后获得可以保持下近似分布不变的属性约简.实验结果证明,本文算法不仅能够大幅减少属性约简时间,而且精度上总体优于NBRS算法和LDNRS算法.

Attribute reduction of variable precision neighborhood rough sets based on improved identification matrix

In this paper, an attribute reduction algorithm is proposed for variable precision neighborhood rough sets, which can greatly reduce the time complexity. The algorithm is based on an improved discernibility matrix. Firstly, consistent and inconsistent decision data is recorded by the matrix at the same time. Then, neighborhood of the sample is computed by binary bit operation. Finally, an attribute reduction that can keep the lower approximation distribution unchanged can be obtained. Experimental results show that this algorithm can greatly reduce the needed time of attribute reduction, and is generally better than NBRS and LDNRS in accuracy.