| Summary: | 碩士 === 立德管理學院 === 應用資訊研究所 === 91 === Data mining has become one of the fast growing areas of research in recent years. In this situation, most enterprises use data mining to improve the productivity. Data mining techniques have become the key factors of enterprises to survive in these years. Besides association rules mining, many researchers exert oneself to develop various methods with transactional databases, time-series databases, spatial information and world wide web. Popular research areas put emphasis on the analysis of the customers purchase patterns. In this thesis, we consider the problem of finding out association rules between items in a large database. Association rule mining is an important model in data mining. We propose an efficient boolean-based algorithm, Multiple-level Boolean algorithm (MB), for exploring the frequent itemsets. The MB algorithm combines the advantages of Multiple-level and Boolean algorithms. Our properly encode transaction table into a set of tables and then we apply Boolean AND and OR operations on the tables to generate frequent itemsets. The combination results in one advantage, which can reduce the number of database scans. In contrast with the Multiple-level algorithm, the MB algorithm has better performance than the Multiple-level algorithm. The Multiple-level algorithm need to scan database once in every pass, that is, if terminate Multiple-level algorithm needs k pass we must scan k times in database. In the same conditions with Multiple-level algorithm, if use MB algorithm just need scan database once in the same direction and the same times of pass.
|
|---|
