基于频率函数循环重计算的属性约简和挖掘算法研究

针对经典HORAFA启发式约简算法在以属性频率为重要启发信息约简时,往往不能获得最优属性约简集的问题,本文提出了基于属性频率函数循环重计算的改进启发式约简和挖掘算法(BRFA算法)。该算法在已约简属性基础上,进行剩余属性频率函数的循环重计算,直至区分矩阵为空,能大大节省决策表的最小约简时间并能得到所有相对约简。通过实例分析和UCI机器学习数据库实验表明,BRFA算法在属性约简和挖掘方面具有较好的性能。     

高可信度最小约简属性启发策略

为提高启发式算法计算最小约简的可信度,基于可辨识矩阵,研究了属性之间存在的吸收、排斥以及互斥等特征,分析其与最小约简的关联,提出了对应的最小约简属性启发策略, 建立了各个特征下属性启发策略的可信度计算模型. 在此基础上,按照可信度排序,形成了一种综合的高可信度最小约简属性启发策略,并给出了具体的约简算法. 理论和实验分析表明,本文策略具有可信度高且可信度可以估计等优点,能有效提升最小约简算法的性能.     

基于分辨矩阵的快速完备约简算法

对分辨矩阵求核过程进行改进与扩展,给出了一种以属性频度作为启发式信息计算最小约简快速完备方法。与其它最小约简算法相比,该算法在很大程度上降低了算法的复杂度,提高了算法的效率。     

一种基于信息熵的信息系统属性约简算法*

粗糙集中找到最小属性约简是一个NP-hard问题,本文根据知识信息熵的定义和性质,定义属性重要性,分析其性质,得出核的求法,给出约简的判定定理,提出了一种基于信息熵的属性约简算法。该算法采用启发式搜索法,先计算属性重要性,求得核,再以核为起点,以属性重要性大小为启发式信息,选择属性,求得最小约简。理论分析和实际计算表明,该算法简明有效。     

基于属性排斥矩阵的约简方法

针对启发式算法难以获得最小约简的问题,研究最小约简约束下属性之间的排斥特性,提出了针对部分最小约简必要条件的属性排斥矩阵.在此基础上,分别结合典型加法类和减法类启发式约简算法提出两种改进的基于属性排斥矩阵的启发式属性约简算法,采用UCI(加州大学欧义分校)机器学习数据集所进行的测试结果表明,属性排斥矩阵能够全面提高启发式属性约简算法的性能,有利于获得最小约简.     

一种完备的最小属性约简方法

为解决粗糙集中的属性约简问题,提出一种完备的最小属性约简方法。将差别矩阵中所有有关属性区分的信息都浓缩进一个差别向量组,计算每个属性在区分2个对象的属性集合中出现的概率,作为属性重要性的启发式信息,建立最小属性约简树,得到属性约简。分析结果表明,该方法可以获得所有的最小属性约简。     

一种基于Rough集理论的属性约简启发式算法

属性约简是知识发现中的关键问题之一．为了能够有效地获取决策表中属性的最小相对约简，在Rough集理论的基础上构造了一个新的算子，将信息论角度定义的属性的重要性作为启发式信息，来描述在决策表中条件属性所提供的知识对决策属性的影响；并采用宽度优先搜索策略，提出了一种新的属性约简启发式算法．以原始条件属性集为起点并结合算子，通过向属性核的递减式逼近，得到属性的最小相对约简．实例分析表明，该算法能有效地对决策表属性进行约简．     

基于属性重要度的启发式特征选取算法

数据库通常包含很多冗余特征,找出重要特征叫做特征提取。本文提出一种基于属性重要度的启发式特征选取算法。该算法以属性重要度为迭代准则得到属性集合的最小约简。     

一种新的启发式粗集决策表属性约简算法

粗集理论通过对原始决策表的约简从而获取规则知识,其核心部分是属性约简.经过约简后的数据更有价值,更能准确地获取知识.文中提出了一种新的启发式属性约简算法,并给出了算法的详细步骤和具体的实验示例.该算法通过不一致计数和互信息增量的计算来衡量属性的重要性,避免了对属性之间随机组合情况的搜索,可以提高求解速度.实验结果表明,相比较于动态约简算法和标准遗传算法,所提出的算法获得的约简属性集更加简洁和高效.     

一种新的不一致决策表的属性约简算法

在基于正域的不一致决策表属性约简算法中,计算正域的算法效率是关键,直接影响到属性约简算法的时间复杂度。针对这一问题,新算法改进了区分矩阵的构造过程,提出了一种有效的在二进制区分矩阵上计算负域的方法,将约简的关键转换为对负域的计算,以属性频率为启发式信息指导属性约简过程。该算法也适用于一致决策表的属性约简。最后,通过实例证明了算法的有效性。     

An enhancement for heuristic attribute reduction algorithm in rough set

Attribute reduction is one of the most important issues in the research of rough set theory. Numerous significance measure based heuristic attribute reduction algorithms have been presented to achieve the optimal reduct. However, how to handle the situation that multiple attributes have equally largest significances is still largely unknown. In this regard, an enhancement for heuristic attribute reduction (EHAR) in rough set is proposed. In some rounds of the process of adding attributes, those that have the same largest significance are not randomly selected, but build attribute combinations and compare their significances. Then the most significant combination rather than a randomly selected single attribute is added into the reduct. With the application of EHAR, two representative heuristic attribute reduction algorithms are improved. Several experiments are used to illustrate the proposed EHAR. The experimental results show that the enhanced algorithms with EHAR have a superior performance in achieving the optimal reduct.     

以准核为起点的启发式属性约简

核是属性约简中的必不可少的部分。通过对核中属性重要程度的差异进行分析,可以发现一些核属性相对于决策的重要度很小,这些属性一定程度上影响了基于约简结果构造的分类器的分类精度。通过将核中一些对决策贡献很小的属性去除,提出了准核的定义,并基于准核构造了一种新的启发式属性约简算法,利用该算法获得的约简中属性数量更少,基于这种约简构造的分类器分类精度更高,实验结果表明了算法的有效性。     

最小相关性最大依赖度属性约简

在经典粗糙集中,基于重要度的决策表属性约简算法只考虑了决策属性与条件属性之间的依赖度,没有考虑约简中条件属性之间的相关性,由此求出的约简中可能依然包含冗余属性。针对这一问题,提出了一种改进算法,它利用最小相关性和最大依赖度准则求决策表属性约简。与基于重要度的决策表属性约简算法相比,本算法求出的约简包含的属性个数少、冗余小。实验结果显示,本算法优于基于重要度的决策表属性约简算法。     

改进的基于简化二进制分辨矩阵的属性约简方法

在基于二进制分辨矩阵的属性约简方法中,删除法即从属性全集中依次删除冗余属性,直至剩余的属性集是一个最小约简.针对传统的基于二进制分辨矩阵的删除法效率较低且得不到最小约简的问题,提出一种改进的二进制分辨矩阵属性约简方法.首先对决策表进行简化,然后给出一种改进的简化二进制分辨矩阵方法;其次通过一个新的属性约简度量方法一次性删除多个属性,并从理论上分析了该方法的可行性;最后通过实验证明了得到的约简结果是最小约简.     

Horizontal decomposition of data table for finding one reduct

Attribute reduction, being one of the most essential tasks in rough set theory, is a challenge for data that does not fit in the available memory. This paper proposes new definitions of attribute reduction using horizontal data decomposition. Algorithms for computing superreduct and subsequently exact reducts of a data table are developed and experimentally verified. In the proposed approach, the size of subtables obtained during the decomposition can be arbitrarily small. Reducts of the subtables are computed independently from one another using any heuristic method for finding one reduct. Compared with standard attribute reduction methods, the proposed approach can produce superreducts that usually inconsiderably differ from an exact reduct. The approach needs comparable time and much less memory to reduce the attribute set. The method proposed for removing unnecessary attributes from superreducts executes relatively fast for bigger databases.     

一种基于粗糙集理论的规则提取方法

规则提取是实现智能信息系统的重要环节,也是一个难点。针对信息系统中的规则提取问题,提出了一种基于粗糙集的研究方法,并对规则提取涉及到的属性约简、属性值约简等问题进行了研究。根据粗糙集中的不可分辨关系建立了可辫识向量,以利用可辨识向量的加法法则运算求得核属性以及属性重要性,然后以核属性为基础、属性重要性为启发信息,求得信息表的一个属性约简。在此基础上,利用条件属性与决策属性之间的对应关系,对信息表中的每条规则通过删除冗余属性值来完成信息表的属性值约简,最终实现规则提取。数值实例和试验表明本算法是有效、可行的。     

一种基于差别矩阵的启发式属性约简算法

为了获得决策系统中更好的相对属性约简,本文提出了一种基于差别矩阵的启发式属性约简算法。该算法以求差别矩阵为基础,不仅考虑了所选择条件属性与决策属性的互信 息,还考虑了其取值的分布情况,从信息论角度定义了一种新的属性重要性度量方法,将其作为启发式信息,最终求得属性约简集。实例表明,算法能够有效地对决策系统进进行约简,获得比较理想的约简结果,同时约简后的决策规则数目较少。     

Attribute reduction: a horizontal data decomposition approach

Ever-growing data generate a need for new solutions to the problem of attribute reduction. Such solutions are required to deal with limited memory capacity and with many computations needed for large data processing. This paper proposes new definitions of attribute reduction using horizontal data decomposition. Algorithms for computing reducts of an information system and decision table are developed and evaluated. In the proposed approach, the size of subtables obtained during the decomposition can be arbitrarily small. The reduct sets of subtables are computed independently from one another using any heuristic method for attribute reduction. Compared with standard attribute reduction methods, the proposed approach can produce the same reducts with less space complexity and with the same or less theoretical time complexity. Experiments conducted under this work show that for information systems with fewer attributes or reducts the time needed for computing the reduct set can be shorten.     

Analysis of alternative objective functions for attribute reduction in complete decision tables

Attribute reduction and reducts are important notions in rough set theory that can preserve discriminatory properties to the highest possible extent similar to the entire set of attributes. In this paper, the relationships among 13 types of alternative objective functions for attribute reduction are systematically analyzed in complete decision tables. For inconsistent and consistent decision tables, it is demonstrated that there are only six and two intrinsically different objective functions for attribute reduction, respectively. Some algorithms have been put forward for minimal attribute reduction according to different objective functions. Through a counterexample, it is shown that heuristic methods cannot always guarantee to produce a minimal reduct. Based on the general definition of discernibility function, a complete algorithm for finding a minimal reduct is proposed. Since it only depends on reasoning mechanisms, it can be applied under any objective function for attribute reduction as long as the corresponding discernibility matrix has been well established.