【作者】 刘伟娜；

【导师】 霍利民；

【作者基本信息】 河北农业大学 ， 农业机械化工程， 2006， 硕士

【摘要】 随着数据库和计算机网络的广泛应用,加之使用先进的自动数据生成和采集工具,人们所拥有的数据量急剧增大,利用信息技术生产和搜集数据的能力也大幅度提高,如何有效地利用和处理信息成为当今世界共同关心的热点课题。由于数据库技术、人工智能和数理统计等技术的不断发展与融合,数据挖掘技术应运而生。数据挖掘是一门新兴的交叉学科,也是现代科学技术相互渗透的必然结果,基本目标就是从大量的数据中提取隐藏的、潜在的和有用的知识和信息。这一技术自20世纪末提出以来,引起了许多专家学者的广泛关注,并应用到金融业、零售业、医疗保健和政府决策等各个领域,取得了良好的社会效益和经济效益,具有广阔的开发前景和应用前景。 贝叶斯网络是概率论与图论相结合的产物,提供了不确定性环境下的知识表示、推理、学习手段,可以完成决策、诊断、预测、分类等任务,因其良好的可理解性和逻辑性成为数据挖掘的重要方法。 本文致力于贝叶斯网络的理论和算法的研究,全文研究了如下几个问题: 1.贝叶斯网络和数据挖掘的结合。贝叶斯网络起源于贝叶斯统计学,数据挖掘本质上具有很强的统计色彩,促成了二者的结合。 2.贝叶斯网络的推理。通过在网络中进行推理,可以得到任意节点间的依赖关系,从而确定数据库中节点所表示事件间的联系,同时对事件的发展进行预测。针对不同的网络,可以采用不同的推理算法,以加快推理速度,提高计算效率。联合树算法以其容易理解,适用范围广等特点成为目前应用最多的精确推理算法。 3.贝叶斯网络的学习。贝叶斯网络的学习是数据挖掘中非常重要的一个环节,是将先验知识和模型评价融入训练数据,获得数据中隐藏的拓扑结构和参数的过程。贝叶斯网络学习分为结构学习和参数学习,其中结构学习是贝叶斯网络学习核心内容。主要对三阶段结构学习算法的原理、实现过程和计算复杂度进行讨论,并利用典型数据库对算法进行验证。参数学习分为:完整数据和不完整数据学习两种,针对不同情况可以采用不同的学习算法,从数据中学习网络的条件概率表。更多还原

【Abstract】 With database and Internet used increasingly and the advanced tools of building and collecting data automatically used widely, the amount of data people hold has been increasing rapidly, and the ability of utilizing information technique and searching data has been improving .How to utilize and deal with the information availably has become a focus all over the world concerned. Database technology, AI and Statistics have been developing and integrating. As a result, Data mining(DM) emerged as the times require. It is a new subject and the necessary result modern science and technology penetrated each other. The basic object of DM is to distill knowledge and information which are concealed, potential and useful from a great deal of data. The technology has been absorbing many experts and used widely in many fields, for example, Finance, Retail, medical treatment and government decision-making since the end of the 20th century. Moreover it has acquired good social benefit and economy benefit, and it will have broad perspective of exploiture and application.Bayesian Network is an outcome probability and map combines and a means which can denote knowledge, reason, learn. It can accomplish decision-making, diagnosis, forecast and classifying, and it has become an important method, which is comprehensible and logical.In this dissertation I dedicate to the research of Bayesian Network’s theory and algorithms .The entire thesis can be divided into three parts.1. Bayesian Networks and Data Mining combine. Because of data mining’s strong statistic characteristic ,as well as Bayesian Networks originating from Bayesian statistics, it causes Bayesian Networks combined with data mining.2. The reasoning technology of Bayesian Networks. Through inferring in networks, we can find out the dependence of two nodes to confirm relation of affairs denoted by nodes, at the same time, to predict the development of these affairs. To different networks, we adopt different inference algorithms to quicken the rate of reasoning and improve calculation efficiency. Because Junction tree algorithm is easy to understand and its range of application is broad , it has been the most used widely among the exact inference algorithms.3. The learning of Bayesian Networks. The learning of Bayesian Networks is an important tache, which combines training data with prior knowledge and model evaluation to acquire the structure hidden in data and parameters. The learning of Bayesian Networksincludes structure learning and parameter learning and structure learning is the core of it. Discuss the learning principles, processes and computational complexity of three phases structure learning algorithm and through typical database validate the algorithm. Parameter learning has two parts: learning from incomplete data and complete data. To different situation, we adopt different learning algorithms to learn condition probability tables (CPT) of Bayesian Network from data.Bayesian principle;Bayesian Network;prior knowledge;probability reasoning;Junction tree algorithm;Three phases algorithm.更多还原