【作者】 刘慧；

【导师】 刘晓东；

【作者基本信息】 大连理工大学 ， 控制理论与控制工程， 2008， 硕士

【摘要】 随着数据库技术的成熟应用和Internet的迅速发展,人们积累的数据量正在以指数级迅速增长。对于这些数据,人们已不满足于传统的查询、统计分析手段,而需要发现更深层次的规律,对决策或科研工作提供更有效的决策支持。于是,分类问题越来越多的出现在人们的生活中,如医学,农业,安全等领域。为了更好的解决实际问题,许多学者试图利用模糊理论,遗传算法,决策树算法,神经网络等方法设计分类器,并且已经初步获得一些成效。随着分类问题的应用越来越广泛,使之有很好的发展前景,研究分类问题变得更有意义。针对现有的AFS(Axiomatic Fuzzy Set)模糊分类器中的不足,本文设计了一种新的基于AFS模糊逻辑的分类器。该分类器的优点主要有:模糊描述简单而且精确;语义明确、易于理解;分类准确率较高:而且,研究表明只用样本属性上的序关系,AFS模糊逻辑分类分析算法也能够获得很高的准确率,因此该算法能够很好的应用到那些样本属性只能用序关系描述而无法用数值描述的数据集,表明本算法具有广范的适用性,而且更多的保持原始数据的信息。将本文设计的分类器应用到UCI(University of California,Irvine)数据库中的三个著名数据集上,做了分类实验,实验得到了较好的效果。通过对实验结果的分析,所提出的分类器是现有基于AFS分类器设计的一个改进,但也发现了本文所设计的算法存在的问题,即多个参数的组合情况易被遗漏和模糊描述语义不易理解的问题。针对这个不足,本文进行了进一步的改进,并将新分类器在四个数据集上做了分类实验,得到了更为理想的结果。实验证明了基于AFS模糊逻辑的分类器的有效性和广泛适用性。最后通过与其他相关的分类方法的分类结果做比较分析,突出了本文提出的分类算法的优点所在。更多还原

【Abstract】 Along with the mature application of data-base technique and the rapid development of the Internet technology, the quantity of data in people’s daily lives is increasing rapidly byexponential. Therefore we need a new method to find the deeper rule which can help provide more efficient decision support since people are not satisfied with conventional searching and statistics analysis method anymore. Then, more and more classification problems appear in our daily lives, such as medicine, agriculture and security field etc. In order to solve the practical problems better, many methods have been tried for designing the classifier, such as the method like fuzzy logic, genetic algorithm, decision tree algorithms and neural network algorithm etc, which have achieved success to some degree, are mainly applied in classifier design. Since the application field is becoming wider and wider, the study on classification problem has brighter future and make the study more meaningful.According to the shorcomings of existing classifier based on AFS fuzzy logic, a new classifier based on AFS (Axiomatic Fuzzy Set) fuzzy logic was designed. The classifier has several advantages, such as short and accurate description, small time consumption,easier sentences for better understanding, relatively high accuracy. Experimental results demonstrated that a high accuracy can be achieved using the proposed classification algorithm by applying only the order relations of the attributes, and the numerical representations of the attribute are not necessary compositions. Thus, this algorithm is widely applicable, and that it can retain more information of the original data.Experiments of classification were done on three famous datasets: wine, iris and WBCD (Wisconsin breast cancer data), which were obtained from database of UCI (University of California, Irvine). After analying experiment results, we found out some problems existed in the AFS Logic algorithm, such as the easy missing problem in case of multi-parameters combinations and that some of the descriptions were difficult to understand. For these problems, modification and improvement of algorithm were done and better performances were achieved from the new experiments on four datasets. At last, through the compare with other algorithms, the advantages of this classification algorithm were shown.更多还原