A Statistical Interestingness Measures for XML based Association Rules

Abstract

Recently mining frequent substructures from XML data has gained a considerable amount of interest. Different methods have been proposed and examined for mining frequent patterns from XML documents efficiently and effectively. While many frequent XML patterns generated are useful and interesting, it is common that a large portion of them is not considered as interesting or significant for the application at hand. In this paper, we present a systematic approach to ascertain whether the discovered XML patterns are significant and not just coincidental associations, and provide a precise statistical approach to support this framework. The proposed strategy combines data mining and statistical measurement techniques to discard the non significant patterns. In this paper we considered the “Prions” database that describes the protein instances stored for Human Prions Protein. The proposed unified framework is applied on this dataset to demonstrate its effectiveness in assessing interestingness of discovered XML patterns by statistical means. When the dataset is used for classification/prediction purposes, the proposed approach will discard non significant XML patterns, without the cost of a reduction in the accuracy of the pattern set as a whole.