An Association Rule Clustering Algorithm Based On K-means And Visualization

Association rule is one of the main content in data mining, and widely used in numerous domains such as marketing, retailing and so on. For massive and high-dimensional datasets, amount of redundant and similar association rules will be generated when the support and confidence threshold are too low, so that the association rules are more difficult to understand and practical apply. In this thesis, association rule clustering algorithm and stellar spectra association rule clusters visualization method have been studied by using k-means idea. The main research works are as follows:(1) An association rule clustering algorithm based on k-means is presented.First, the redundant association rule is redefined, then its deleting method is proposed. Second, a new similarity measure among the rules is defined according to the structure characteristics between antecedent and consequent of association rules. Third, the initial cluster points are selected by using largest triangle method, and association rules without redundant rules are clustered by using K-means' s idea, then similar rules are put into one cluster. In the end,experiments on stellar spectra data and simulated datasets verify that this algorithm could help users quickly seek useful association rules.(2) A stellar spectra association rule clusters visualization method based on parallel coordinate is presented. First, the attribute dimensions(horizontal) are defined by properties in stellar spectra association rule clusters; Second attribute interval(vertical) are defined by the number of each attribute's characteristics;Then, specific dimension or specific attributes' regularities are displayed by using brush and increase or decrease dimensionality technique. Finally,visualization experiments on stellar spectra association rule clusters, which verify it can effectively improve the understandability of association rules.