模糊决策粗糙集代价敏感属性约简研究

针对决策中普遍存在的代价问题,在模糊理论和决策粗糙集的基础上,对其代价敏感属性约简方法进行了研究。在模糊决策粗糙集属性约简中引入了包含误分类代价和测试代价的总代价。因此约简的目标不再只是考虑正域的大小,而是寻找使得总代价最小的最优属性子集。提出了一种模糊决策粗糙集代价敏感属性约简(COSAR)算法,该算法采用启发式方法搜索最优属性子集。给出了算法的步骤,并将该算法与已有的模糊粗决策粗糙集属性快速约简(QuickReduct)算法进行了性能对比。实验结果表明,COSAR算法比QuickReduct算法具有更强的属性约简能力、更低的分类总代价、更短的运行时间,且随着测试样本的增加,分类总代价差值也越来越大。

Study on Cost Sensitive Attribute Reduction for Fuzzy Decision Theoretic Rough Sets

Aiming at the cost problem that generally exists in desicion-making,on the basis of fuzzy theory and decision theoretic rough sets,we studied the method of cost sensitive attribute reduction.We introduced the total cost including misclassification cost and test cost into attribute reduction for fuzzy decision theoretic rough sets (FDTRS).Thus the target of reduction is not only to considere the size of positive region,but also to find the optimal subset of attributes with the minimum total cost.We proposed a cost sensitive attribute reduction (named as COSAR) algorithm for FDTRS.The algorithm uses a heuristic method to search the optimal subset.We provided the procedure of the algorithm and compared the performance of the algorithm with the existing FDTRS attribute reduction algorithm,called QuickReduct.The experimental results show that COSAR algorithm has stronger attribute reduction capability,lower total classification cost,shorter running time than QuickReduct algorithm,and with the increasing of test samples,the difference of total classification cost between two methods is growing larger.