Multidimensional Multi-granularities Data Mining for Discover Association Rule

Abstract

Data Mining is one of the most significant tools for discovering association patterns for many knowledge domains. Yet, there are deficits of current data-mining techniques, i.e.: 1) current methods are based on plane-mining using pre-defined schemata so that a re-scanning of the entire database is required whenever new attributes are added. 2) An association rule may be true on a certain granularity but false on a smaller ones and vise verse. 3) Existing methods can only find either frequent rules or infrequent rules, but not both at the same time. This paper proposes a novel algorithm alone with a data structure that together solves the above weaknesses at the same time. Thus, the proposed approach can improve the efficiency and effectiveness of related data mining approach. By means of the data structure, we construct a forest of concept taxonomies which can be applied for representing the knowledge space. On top of the concept taxonomies, the data mining is developed as a compound process to find the large-itemsets, to generate, to update and to output the association patterns that can represent the composition of various taxonomies. This paper also derived a set of benchmarks to demonstrate the level of efficiency and effectiveness of the data mining algorithm. Last but not least, this paper presents the experimental results with respect to efficiency, scalability, information loss, etc. of the proposed approach to prove its advantages.