【作者】 李凡；

【导师】 白素怀；

【作者基本信息】 西安电子科技大学 ， 情报学， 2002， 硕士

【摘要】 本文以数据挖掘中的数据分类算法为主要研究对象。对决策树分类、贝叶斯网络和连续属性的离散化问题进行了的研究，实现了多种分类算法。在此基础上，利用最相邻子集筛选原理提出了两种决策树的优化算法LDT+和SubBagging,提高了原有分类算法的分类准确度。在来自不同领域的数据集上进行了大量的分类实验，分析和比较了多种决策树分类算法的分类性能和对不同数据的适应性。 研究了贝叶斯网络的拓扑结构学习、参数学习和网络评估等问题，在此基础上设计实现了一种基于属性相关性分析的贝叶斯网络学习算法。同时，对分类算法中连续属性的离散化问题进行了研究。通过具体实验，分析和比较了层次聚类法、递归最小熵法和One-Rule等离散化方法的性能特点。 在数据挖掘算法研究的基础上设计并实现了一个用于数据分类的数据挖掘系统——InfDM。在系统设计中提出了一种数据挖掘系统和专家系统协同工作的设计思想和实现方案。实践证实InfDM有较强的数据分类能力、可交互性和可扩展性。 本论文得到总装备部“交互式在线信息服务技术研究”这一技术基础项目的支持。更多还原

【Abstract】 In this paper the algorithm of the data classification in the data mining is regarded as the main research target. A further study has been made about decision tree classification, Bayesian network, and discretization of conntinuous attributes, at the same time many kinds of classfication algorithms have been achieved. Based upon the algorithms, by using the pricinple of nearest neighbor subset selection, two kinds of optimized algorithms which are LDT+ and SubBagging of the decision tree are present, which improves the accuracy of the origined classification algorithms. We also make plenty of classification experiments with data sets from various of different fields, and then analyse and compare the classification capacity of several decision tree classification algorithms and the adaptability to different datas.We make some further study on some problems, such as the learing of structure and parameters of Bayesian Network, network estimate and so on , on the basis of which a kind of learning method of Bayesian network based on the attribute relativity analyse is achieved.We also study the problem of discretization of conntinuous attributes in data mining. Through some specific experiments, we analyse and compare the characters of some discretization methods such as hierarchical clustering analysis, recursive minimal entropy method, and One-Rule.A data mining system-In/DM which applys to data classification is designedand realized based on the study on the algorithms above. During the course, a designing idea and a realization project about the combination of the data mining sysytem and the expert system are present. In/DM has the great ablity of data classification, interactivity and expansibility, which is proved by experiments.The paper is supported by one of the technical basis projects of General Equipment Department of L.A.M with the name "Technical Reasearch on Interactive Online Information Service".更多还原