Summary: | 碩士 === 國立臺灣科技大學 === 電子工程系 === 90 === Mining frequent patterns in large database is an important issue in Data Mining. Extensive study of mining frequent patterns have been focused on Memory-based approach in recent years. In such environment, effective reduction in memory overhead becomes a critical problem. In this thesis, we investigate a dynamic sharing structure and to support mining frequent patterns without candidates generation- and-test. The major distinct feature of the proposed algorithm is small memory requirement in comparison with other approaches, especially when a large number of items appear in each transaction. Moreover, we propose a new counting frequent pattern scheme, which effectively prune the search space and enhance the execution speed. To evaluate the performance of the proposed algorithm, we have conducted an extensive performance study. The results indicate that the proposed algorithm achieves less space requirements and faster execution time than H-Mine algorithm.
|