期刊名称：International Journal of Computer Science and Information Technologies
电子版ISSN：0975-9646
出版年度：2017
卷号：8
期号：2
页码：232-236
出版社：TechScience Publications
摘要：The association rule mining has been very useful in manybusiness applications, such as market analysis, web data analysis,decision making, knowing customer purchase behavior etc. In almostall transactional databases, new transactions are added with time. Soan efficient algorithm is required to be developed in order to avoid rescanningof old datasets. Incremental mining [18] deals withgenerating association rules based on available knowledge (obtainedfrom mining of previously stored databases) and incrementeddatabases, without scanning the previously mined databases again. Inthis paper a novel approach of using horizontal and vertical databaselayout, a new representation of transactional database, has beenproposed (modified version of Inverted Matrix [17]) to mine largedatabase incrementally. One of the major advantage of this approachis that it generates the data structure in a single scan of the databaseand whenever new database is added incrementally, the generatedstructure is updated to consider the effect of incremented database. Inthis paper the Modified Inverted Matrix is distributed amongstparallel nodes. Frequent item from the modified inverted matrix isassigned to parallel nodes in alternate splitting fashion. In parallelimplementation, a Co-Occurrence Frequent Item (COFI) tree [17] forassigning frequent item is generated by the parallel nodes. Miningprocess is accomplished by all nodes which generate all frequent itemsin which the assigned items are participated. Here, lesscommunication is required amongst the master node and all parallelnode to generate all frequent item sets. Moreover, one of theadditional advantage is that the algorithm still responds correctlywithout the need for changing the data structure, whenever desiredsupport value changes. In this paper we provide the theoretical proofof concept for our proposed approach.
