Abstract: In the feature reduction algorithms designed by the discernibility matrix idea，since there are lots of repeat and unnecessary elements in the discernibility matrix，which not only cost a mass of memory space，but also waste plenty of computer time in feature reduction.In order to improve the efficiency of such feature reduction algorithm，by considering the idea of FP tree，a novel data structure IFP_Tree（improved frequent pattern tree） is proposed，which can get rid of the repeat elements and unnecessary elements in the discernibility matrix completely.In this way，it can not only reduce a great deal of memory space，but also enhance the efficiency of feature reduction algorithm greatly.Then，a new quick and efficient feature reduction algorithm is designed based on IFP_tree.Finally，an example is used to illustrate the validity of the new algorithm.

Key words: rough set, discernibility matrix, feature reduction, improved frequent pattern tree

摘要： 在用差别矩阵思想设计的属性约简算法中，由于差别矩阵存在大量重复和无用的差别元素，不仅占用大量的存储空间，而且浪费属性约简的计算时间。为提高这种属性约简算法的效率，结合FP树（频繁模式树）的思想，给出一种新型的数据结构——改进的FP树（IFP_Tree）。改进的FP树可以完全删除差别矩阵中所有重复的差别元素，也可以完全删除无用的差别元素。不但减少了大量的存储空间，还大大提高了属性约简算法的效率。用IFP树设计一种新的快速属性约简算法。实例说明了该算法的有效性。

关键词: 粗糙集, 差别矩阵, 属性约简, 改进的FP树