基于模糊粗糙集依赖度的两步属性约简方法

Two-step Attribute Reduction Method Based on Fuzzy Rough Sets Dependency

为获取连续属性值数据集的最小属性子集,提出了一种两步约简方法.该方法以模糊粗糙集模型为基础,将描述条件属性和决策属性依赖关系的模糊依赖度概念进行了扩展,使其能对条件属性之间的依赖关系进行度量,利用属性与类别之间的依赖度选出候选属性集,然后根据单个属性与类别和属性之间的依赖度对候选属性集进行约简.仿真结果表明,该方法在有效降低属性维数的同时一定程度上保证了分类正确率.

To acquire the minimum attribute reduction of the dataset with continual attribute value, a two- step reduction method is proposed based on fuzzy rough sets. The concept of dependency is extended and the dependency relationship between condition attributes can be measured on the basis of the extended concept. The candidate attributes are first selected by calculating the attribute importance. Then, the candidate attributes are reduced using the dependency between attributes and of the single attribute importance. The redundant attribute is reduced via this operation. Experimental results show that the strategy can efficiently reduce the attribute dimension without sacrificing the classification accuracy.