Mining Maximal Frequent Patterns In A Unidirectional FP-tree

Data mining is a young multidisciplinary field, drawing work from areas including database technology, machine learning, statistics, pattern recognition, neural networks, artificial intelligence, and data visualization etc. Nowadays, it is one of the most active and exciting areas of the database research community.Association rule is one of the important research areas in data mining. Its goal is to discover previously unknown, interesting relationships among attributes from large databases. The first step for mining association rules is mining frequent itemset which is also the key step can influence the total mining performance. Therefore, in this paper we put the emphasis on mining frequent itemset. The main research is as follows:1.The two algorithms for mining frequent itemset are thoroughly studied. One is the classical algorithm, FP-growth, which leads the way to mining frequent patterns without candidate generation. Another is the algorithm proposed by Fan Ming for mining frequent itemset in a unidirectional FP-Tree, which doesn't generate conditional FP-trees in mining process. We comparatively analyze the similiarities and differences of FP-Tree and unidirectional FP-Tree, summarize the problems on FP-growth and the advantages in Fan Ming's algorithm.2. Based on foregoing research and compared with the algorithm FP-Max, we propose the algorithm Unid_FP-Max for mining maximal frequent itemset in a unidirectional FP-Tree. The Unid_FP-Max is a depth-first search algorithm. Algorithm analysis and experiments show that the Unid_FP-Max consumes less space and time than the FP-Max for the dense data.3.Referred to the algorithm CLOSET, the algorithm Unid_FP-FCI is designed for mining frequent closed itemset in a unidirectional FP-Tree, which is a depth-first search algorithm. Rough analysis indicate that the Unid_FP-FCI is more efficient than the CLOSET for the dense data.