基于容差关系知识依赖的属性约简算法研究

Research on attribute reduction algorithms based on knowledge dependence by tolerance relation

为了提高不完备信息系统中的属性约简能力,文中基于容差关系研究了知识粒度、知识依赖度等概念,给出了一种新的知识粒度定义,研究了两种依赖度的性质,并通过例子加以验证.通过引入知识粒度的概念,对信息系统中属性的重要度进行了定义;并以属性重要度作为启发式信息提出了属性约简的两种算法,一个从核属性集出发,采用自底向上的方法,另一个从整个属性集出发,采用自顶向下的方法.在5组UCI数据集上的实验结果表明,从约简个数来看,文中算法与其他算法相比,属性个数少于或等于其他算法结果;从分类精度来看,从核属性集出发的算法精度更高.最后通过实验验证了文中算法的正确性和可行性.

In order to improve the ability of attribute reduction in incomplete information systems,we study the concepts of knowledge granularity and knowledge dependence based on tolerance relation,and give a new definition of knowledge granularity.Then the properties of two kinds of dependencies are studied and verified by examples.By introducing the concept of knowledge granularity,the importance of attributes in the information system is defined,and two algorithms of attribute reduction are proposed by using attribute importance as heuristic information.One starts from the core attribute set using the bottom-up method,and the other from the whole attribute set using the top-down method.Experimental results on 5 sets of UCI datasets show that,from the reduction of number,the algorithm is less than or equal to the result of the number of attributes compared with other algorithms,from the perspective of classification accuracy,the algorithm from the kernel attribute set is better.The correctness and feasibility of the algorithm are verified by experiments.