基于下近似分布的变精度邻域粗糙集属性约简算法

Attribute Reduction of Variable Precision Neighborhood Rough Set Based on Lower Approximate Distribution

利用邻域粗糙集处理数值型数据,可以解决经典粗糙集不能直接处理数值型数据的问题,改进后的变精度邻域粗糙集可以增强抗噪声的能力。但变精度邻域粗糙集的属性约简有不同于邻域粗糙集的特性,需要考虑每个决策类的下近似分布。文中提出可以遵循平均错误率来约简属性,减少计算规模。实验证明,使用UCI数据集与其它算法进行了比较,该算法可以获得理想的结果。

Problems of numerical data that can＇t be dealt with directly by classical rough sets can be solved by using neighborhood rough sets. Based on this concept,the ability to resist noise can be enhanced by improved variable precision neighborhood rough sets. However,the attribute reduction of variable precision neighborhood rough sets is different from that of neighborhood rough sets,and the lower approximation distribution of each decision class needs to be considered. A model was proposed in this paper,for which the average error rate can be used to help reduce attributes and downsize the computational scale. Experiments show that satisfactory results can be reached when this algorithm is compared with other algorithms by UCI data sets.