A Study of Association Rules Based on Trees

• Main Authors: Wei-Chi Shiu, 許瑋琪
• Other Authors: Yuh-Jiuan Tsay
• Format: Others
• Language: zh-TW
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/70703158178898004404

碩士 === 國立屏東科技大學 === 資訊管理系 === 94 === The purpose of this thesis is to overcome the weaknesses of association rules, such as requiring repeated passes over the database, and generating a large number of the candidate itemsets and conditional FP-trees. We present a new algorithm named Large-Item Based Trees Association Rule (L-trees). The first step of L-trees is to cluster the database into N (N is number of L1) tables, and delete the Item that count is smaller than support for each table, then construct the L-trees from the table which is pruned. Finally, mine the L-trees recursively to find all frequent itemsets. L-trees is only to scan database twice and doesn’t sort the transactions. The major difference between the FP-growth and COFI-tree is that it needn’t construct the FP-tree in advance, but constructing the L-trees from the table which is pruned immediately and mining all frequent itemsets. This experiments show that L-trees algorithm outperforms FP-growth algorithm, a well-known and widely used association rule.