决策树技术研究与应用.pdf

决策树技术研究与应用 摘 要 数据挖掘是当前涉及统计学、人工智能、数据库等学科的热门研究领域，是从数据中提取人们 感兴趣的、潜在的、可用的知识，并表示成用户可理解的形式。 分类是数据挖掘的一个重要分支，分类能找出描述数据类或概念的模型，以便能使用模型预测 类来标记未知的对象类。基于决策树的分类技术，与其他分类技术相比，具有很多优点：决策树更 容易被用户理解；生成决策树的效率更高、更适合大训练集；决策树的生成算法不需要除训练集以 外的额外信息；可以提供更好的精确度。当然决策树也有缺点：一方面，它无法删除带噪声的不相 关的属性；另一方面，大多数决策树被限制在每个节点上只检验单个属性。 本文针对分类中条件属性较多时，其中的个别条件属性对分类不具有影响的情况，讨论了使用 决策树技术的简约办法，为更有效的进行数据挖掘在方法层面上提供了支持。 关键词： 数据挖掘 决策树 分类技术 条件属性 I Application Research of the Decision Trees Abstract At present, Date mining is a hot research topic involving statistics, artificial intelligence and database. Data mining extracts the useful and potential knowledge that we are interested in, and expresses it in a comprehensible form. Classification is one crucial branch of data mining, which can find out the models of data types or definitions to predict through models the class-unmarked object types. The thesis emphasizes on the classification technique based on decision trees. Decision trees display good accuracy as compared to other techniques: more comprehensible to the users, more efficient in creating a decision tree and more suitable for large training sets. But decision trees also have some limitation. One hand, it can not delete irrelevant attributes; On the other hand, most decision trees can test only one attribute on each node. In addition, the simplification approach of decision trees is given when there are many conditional attributes with useless individual attribute. It provides the support on the approach of the data mining. At the end of the thesis, data are collected an