摘要
针对关联分类算法面对大数据集事务库时的候选集庞大、难以生成挖掘数据结构和高资源消耗等问题,提出了一种基于投影数据库的改进单向COFI-tree频繁项集生成算法,结合相关性度量等剪枝操作提取高效的分类规则,通过分治数据库有效降低整个数据库对资源的需求,减小对频繁项集的搜索空间和非频繁项集的数量,从而实现对频繁项集生成的优化过程.实验结果表明该算法通过生成初始投影数据库,并利用单向COFI-tree挖掘频繁项集的时间远小于同类算法对数据集进行直接挖掘,为大数据集的关联分类挖掘提供了一种新的解决途径.
For the purpose of solving the associative classification algorithm in large data transaction datasets when the candidate set is large, difficult to generate data mining structure and high consumption of resources, a new frequent itemsets generation algorithm based on projected database and improved single direction COFI-tree is proposed, combined with the correlation measure to get the ef- ficient class association rules. The whole large data transaction datasets is divided into several projected databases which reduced the demand for resources, and decreases the traversal space and the sum of infrequent itemsets, so as to realize the optimization process to generate frequent itemsets. Experimental results show that the algorithm by generating the initial projected database, and mining fre- quent itemsets time is far less than the same kind of algorithms, and provides a new solution for associative classification of mining large dataset.
出处
《小型微型计算机系统》
CSCD
北大核心
2014年第4期791-796,共6页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61163057)资助
广西可信软件重点实验室项目(kx201111)资助
广西教育厅科研项目(201012MS088)资助
