【作者】 邹先霞；

【导师】 魏长华；

【作者基本信息】 华中师范大学 ， 计算机应用， 2001， 硕士

【摘要】 真实的物理世界是不确定的世界，不确定性信息和知识在人们工作、生活等各方面随处可见。为适应诸如科学研究、金融投资、市场营销及预测等这些具有不确定性领域的需要，知识发现和数据挖掘已成为当前一个具有重要理论和实用价值的研究课题，并得到国际学术界尤其是人工智能工作者的广泛关注。本文在深入学习并熟悉数据挖掘的基本特点和相关技术的基础上，充分分析了国际上数据挖掘方面已取得的研究成果及其存在的不足，对知识发现和数据挖掘开展研究。 论文首先介绍了国内外关于知识发现和数据挖掘的发展动态、数据挖掘的基本概念和一般处理过程。我们通过对波兰华沙理工大学计算机研究所的Zdzislaw Pawlak教授提出的标准粗集理论的研究，发现在标准粗集理论中的等价关系过于严格，使之在实际应用中难于实现。基于此，我们提出了容错粗集理论的概念，用容错关系代替了严格的相等关系，并提出了与标准粗集理论相对应的基本概念，如容错上近似、容错下近似、容错关系集等。在构造容错关系时，我们提出采用遗传算法来获取属性集上的最优阈值组合，使等价类的划分更接近于真实。最后，我们通过对知识发现和数据挖掘的深入研究，发现它主要是对数据进行深入分析，其过程缺少了用户的参与，而很多的用户往往希望能按他们所期望的目标进行挖掘。由此，提出了基于容错粗集理论的联机分析挖掘技术的理论，使得用户在挖掘的过程中，动态地提出挖掘的对象和要求，系统不断地调整挖掘算法，实施再挖掘。 本文的主要成果有： （1）深入研究了KDD与粗集理论的基本概念和方法，提出了用容错关系代替标准粗集理论的相等关系。 （2）提出了容错粗集理论的基本概念。 （3）提出了基于容错粗集理论的数据挖掘算法。 （4）提出了将容错粗集理论的挖掘对象建立在数据仓库上，与OLAP技术相结合，即OLAM技术，并证明这是可行的，因为基于容错粗集理论的数据挖掘技术不再进行初始量化。 （5）提出一个OLAM的框架，为系统的进一步完善奠定理论基础。更多还原

【Abstract】 The world is a place of uncertainty in that uncertain information and knowledge is congested everywhere. In response to the rapid development of computer technology and extensive requirements of applications such as scientific research, finance investment, market management and prediction, etc.. Knowledge discovery in database and data mining is becoming a highlighted research topic on account of its important theoretical and practical value. Upon investigating into the research achievements and shortcomings of existing in knowledge discovery in database and data mining, this thesis has developed the research.First we introduced the developing trends of home and abroad regarding knowledge discovery in database and data mining, basic concepts and typical processing. Then we make thorough research into general Rough Set Theory (RST), which was introduced by Professor Zdzislaw Pawlak. We find that the equivalent relation is too restricted to be achieved in the reality. So we advance tolerance rough set theory (TRST) and the concept corresponding to it, for example, tolerance upper approximation, tolerance lower approximation, tolerance relation sets and so on. We also construct the best threshold combination with genetic algorithm. The method improves the equality of the classification and the precision. Finally we find that since knowledge discovery in database mainly makes analysis to data, its process is completely automatic in defect of the interference of the user. But some users hope the data mining process done according to their needs. We know that on-line analysis processing has the virtues of mutual capability and speedy response. So we bring forward the method of on-line analysis mining (OLAM) based on TRST, with which users can take part in the process of the mining and dynamically decide the objects and the task of mining.The main fruits of this paper are:(1) The basic concepts and methods of KDD and RST are studied intensively, and Tolerance relation takes the place of the equivalent relation.(2) The basic concepts of TRST are proposed.II(3)The algorithm of data mining based on TRST is given.(4)The object of data mining based on TRST is built on the data warehouse, and OLAP is joined with data mining based on TRST, which is feasible because data mining based on TRST does not initialize the quantification.(5)The framework of OLAM is proposed, which can improve Atmosphere Environment Forecast System.更多还原