| Association analysis, one of the popular data mining methods, is to discover hidden relationships among items in a huge dataset. The process includes generations of frequent itemsets and association rules. The frequent itemset generation is the performance bottleneck in time and space in addition to possible rare itemset and cross-support pattern problems. Among all the algorithms which have been developed to improve the performances, the FP-growth tree family is most widely adopted: FP-tree converts a dataset into a compressed prefix tree to save I/O time and CFP-tree includes multiple minimum supports to remove the rare itemsets problem. However, they all suffer from running time and the cross-support patterns. This thesis proposes a novel technique, CFPh-growth tree, running in less time/space for association analysis without generating cross-support patterns and rare itemset problems. The experiment results show that the technique achieves these goals and drastically reduces the running time and space. |
