Research On Association Rules Mining Algorithm Based On Closed Pattern

Data Mining is a hot research topic in database and artificial intelligence fields to find the useful knowledge hidden in the vast data. Association rules are an important part of data mining, which could describe the relationships between multi items that may be useful in market analysis and business decision-making.Mining closed patterns can produce much smaller equivalent result sets than mining frequent patterns. Now, there have been many algorithms to mining frequent closed patterns. But they can't provide a best approach to mining closed patterns. They must conserve the closed patterns in memory or traverse the diff-set and compute the closure relation, so they can't get the high efficiency and good scalability. The number of produced rules is very large, and many of them are useless to users. In order to resolve these problems, we do some researches on association rules mining based on closed patterns.This paper first introduces some theory knowledge, such as the background knowledge of data mining and association rules. Next we interpret the definition of closed patterns which provide a minimal representation of itemsets without losing support information. After that, based on the theoretical framework of Formal Concept Analysis and Iceberg Concept Lattice, we discuss how to extract relatively small bases for association rules from which all rules can be deduced. Then, a new algorithm for mining frequent closed patterns based on frequent pattern tree, FCIM algorithm is presented. The algorithm is in accordance with the divide-and-conquer method. We advance a concept of equal-child in the algorithm, with it we could utilize the foregone information to reduce redundant complicated operations according to the order from bottom to top, and then find out every frequent closed pattern without duplicates generation and memory conservation. We also take two measures to predict and extract the closed patterns as soon as possible. It helps to solve a number of problems that exist in the current algorithms, such as high memory request, repetitious I/O accessing, and so on. At last, we describe our algorithm and list an example.The experiments prove the correctness and the better performance of this algorithm. After comparison with CLOSET+ and analysis in our experiments, we can conclude that the algorithm is efficient and practical.