摘要
属性约简是粗糙集理论的重要研究内容之一。分布约简保证约简前后每个对象的概率分布保持不变,即保证每条规则的置信度在约简前后不发生改变。实际应用中,人们往往更加关注可信度较高或较低的规则。因此,在本文中引入了广义分布保持属性约简,该属性约简可以保证规则的置信度P(P∈[0,α]或[β,1])在约简前后不变。同时,给出了广义分布保持属性约简的判定方法与基于差别矩阵的广义分布保持属性约简算法,深入讨论了几种特殊情形下的广义分布保持约简。最后,在4个UCI数据集上进行的实验分析表明,几种特殊情形下的广义分布保持属性约简可退化为已有的一些属性约简,且在不同置信区间下求得的广义分布保持属性约简存在包含关系,验证了相关结论的正确性。
Attribute reduction is a pertinent issue in rough set theory. Distribution reduction ensures that the probability distribution of each target does not change before and after reduction; i.e.,it ensures that the confidence of every rule remains unchanged before and after reduction. In actual applications,people are often interested in rules that have higher or lower confidences. Thus,attribute reduction based on generalized distribution preservation is proposed in this paper. Confidences in [0,α]or [β,1]were unchanged using the proposed technique. We also propose judgment methods for generalized-distribution-preservation attribute reduction and investigate the generalized attribute-reduction algorithm based on a discernibility matrix. Some special cases with respect to generalized-distribution-preservation attribute reduction are discussed in depth. Finally,experiments on four data sets downloaded from UCI show that some special cases with respect to generalized distribution preservation reduction could degenerate into some existing attribute reductions and inclusion relations exist in generalized distribution preservation attribute reduction under different confidence intervals,verifying the correctness of the relevant conclusions.
出处
《智能系统学报》
CSCD
北大核心
2017年第3期377-385,共9页
CAAI Transactions on Intelligent Systems
基金
国家自然科学基金项目(61403329
61572418
61502410
61572419)
山东省自然科学基金项目(ZR2013FQ020
ZR2015PF 010)
山东省高等学校科技计划项目(J15LN09
116LN17)
关键词
分布保持
属性约简
粗糙集
概率分布
差别矩阵
distribution preservation
attribute reduction
rough sets
probability distribution
discernibility matrix
