Mining frequent pattern in transaction database, time series databases, and many other kinds of databases have been studied popularly in data mining research. Most of the previous studies adopt Apriori like candidat...Mining frequent pattern in transaction database, time series databases, and many other kinds of databases have been studied popularly in data mining research. Most of the previous studies adopt Apriori like candidate set generation and test approach. However, candidate set generation is very costly. Han J. proposed a novel algorithm FP growth that could generate frequent pattern without candidate set. Based on the analysis of the algorithm FP growth, this paper proposes a concept of equivalent FP tree and proposes an improved algorithm, denoted as FP growth * , which is much faster in speed, and easy to realize. FP growth * adopts a modified structure of FP tree and header table, and only generates a header table in each recursive operation and projects the tree to the original FP tree. The two algorithms get the same frequent pattern set in the same transaction database, but the performance study on computer shows that the speed of the improved algorithm, FP growth * , is at least two times as fast as that of FP growth.展开更多
基金theFundoftheNationalManagementBureauofTraditionalChineseMedicine(No .2 0 0 0 J P 5 4 )
文摘Mining frequent pattern in transaction database, time series databases, and many other kinds of databases have been studied popularly in data mining research. Most of the previous studies adopt Apriori like candidate set generation and test approach. However, candidate set generation is very costly. Han J. proposed a novel algorithm FP growth that could generate frequent pattern without candidate set. Based on the analysis of the algorithm FP growth, this paper proposes a concept of equivalent FP tree and proposes an improved algorithm, denoted as FP growth * , which is much faster in speed, and easy to realize. FP growth * adopts a modified structure of FP tree and header table, and only generates a header table in each recursive operation and projects the tree to the original FP tree. The two algorithms get the same frequent pattern set in the same transaction database, but the performance study on computer shows that the speed of the improved algorithm, FP growth * , is at least two times as fast as that of FP growth.
文摘选择性集成通过选择部分基分类器参与集成,从而提高集成分类器的泛化能力,降低预测开销.但已有的选择性集成算法普遍耗时较长,将数据挖掘的技术应用于选择性集成,提出一种基于FP-Tree(frequent pattern tree)的快速选择性集成算法:CPM-EP(coverage based pattern mining for ensemble pruning).该算法将基分类器对校验样本集的分类结果组织成一个事务数据库,从而使选择性集成问题可转化为对事务数据集的处理问题.针对所有可能的集成分类器大小,CPM-EP算法首先得到一个精简的事务数据库,并创建一棵FP-Tree树保存其内容;然后,基于该FP-Tree获得相应大小的集成分类器.在获得的所有集成分类器中,对校验样本集预测精度最高的集成分类器即为算法的输出.实验结果表明,CPM-EP算法以很低的计算开销获得优越的泛化能力,其分类器选择时间约为GASEN的1/19以及Forward-Selection的1/8,其泛化能力显著优于参与比较的其他方法,而且产生的集成分类器具有较少的基分类器.
