基于K-means的关联规则聚类分析与可视化

发布时间：2018-04-23 22:14

本文选题：关联规则 + 聚类算法　；参考：《太原科技大学》2016年硕士论文

【摘要】：关联规则是数据挖掘的主要研究内容之一,并已广泛地应用在市场营销、零售等众多领域。然而针对高维海量数据集,尤其当支持度和置信度阈值太低时,将生成大量冗余和相似的关联规则,对关联规则的理解和实际应用造成了较多困难。本文采用K-means思想,对关联规则聚类算法与恒星光谱关联规则聚类簇的可视化方法进行了研究,主要成果如下:(1)给出了一种基于K-means的关联规则聚类算法。该算法首先重新定义了冗余关联规则,并给出了一种删除冗余关联规则的方法;然后根据关联规则前件与后件的结构特性,定义了一种新的规则间相似性度量;其次采用最大三角形方法选取聚类的初始点,并利用K-means思想,对删除冗余后的关联规则进行聚类分析,将相似的关联规则归为一类;最后采用恒星光谱数据和人工数据集,实验验证该算法能够帮助用户快速地找到有用的关联规则。(2)给出了一种基于平行坐标的恒星光谱关联规则簇可视化方法。该方法首先根据恒星光谱关联规则簇中的属性,对属性维(横坐标)进行定义;然后根据各属性的特征个数对属性区间(纵坐标)定义;其次,采用刷技术、增减维数来显示特定维数或者特定属性时的规律。最后,采用恒星光谱关联规则聚类集进行可视化,实验验证该方法有效地提高了关联规则的可理解性。
[Abstract]:Association rules are one of the main research contents of data mining and have been widely used in many fields, such as marketing, retail and so on. However, a large number of redundant and similar association rules will be generated for the high dimensional mass data sets, especially when the support and confidence threshold are too low, which causes more difficulties in the understanding and practical application of association rules. It is difficult to use the K-means idea to study the visualization method of association rule clustering algorithm and stellar spectral association rule clustering cluster. The main results are as follows: (1) a clustering algorithm based on K-means is given. The algorithm first redefines the redundant association rules, and gives a redundant association rule. And then, according to the structural characteristics of the association rules and the post parts, a new measure of the similarity between rules is defined. Secondly, the initial point of the cluster is selected by the maximum triangle method, and the K-means idea is used to cluster analysis of the redundant association rules, and the similar association rules are classified as a class. Finally, the constant star light is used. The algorithm can help users to quickly find useful association rules. (2) a visualization method of star spectral association rules based on parallel coordinates is given. Firstly, the attribute dimension (abscissa) is defined according to the attributes of the star spectral association rules cluster; and then according to each of them, The characteristic number of attributes is defined on the attribute interval (ordinate); secondly, the brush technique is used to increase and decrease the dimension to display the specific dimension or specific property. Finally, the stellar spectral association rule clustering set is used to visualize it. The experiment proves that the method effectively improves the comprehensibility of the association rules.

【学位授予单位】：太原科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP311.13

【相似文献】