Font Size: a A A

Research Of Classification Algorithm Based On Association Rule Mining

Posted on:2013-03-29Degree:MasterType:Thesis
Country:ChinaCandidate:L S XuFull Text:PDF
GTID:2248330362972207Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
With the rapid development of information society, the importance of data mining in allfields is more and more prominent. In the area of data mining, classification is an importantanalytical method, and association rule mining is an important research direction. As they aretwo highly active research areas in data mining, they have similarities in mining item setswith strong relevance. Therefore, the combination of the two techniques that apply miningassociation rules in the task of classification opens a new journey for dataclassification-associative classification.Associative classification is essential classification based on association rules, which notonly reflects the application characteristics of knowledge-classification and prediction, butalso embodies the inherent associated characteristics of knowledge. The differences betweenthe associative classification methods are mainly reflected in two aspects: the method used inmining frequent item sets and analyzing the mined rules for classification.On the base of analyzing and comparing both strengths and weaknesses of the existingassociative classification algorithm, this paper presents an associative classification algorithmbased on P-Trie tree, named CARPT. This algorithm uses a vertical data format to compressand store the original database, which reduces the number of database scanning and makes thesupport counting convenient and improves the efficiency of the algorithm. The algorithmregards a frequent item set as a string and uses P-Trie tree to store the frequent information, tomine class association rules. It removes the frequent item sets that cannot generate frequentrules directly by adding support count for class labels of frequent items during theconstruction of P-Trie tree. This strategy is equivalent to pre-pruning of P-Trie tree; it caneffectively reduce the number of nodes of the P-Trie tree and the round-trip time of miningprocess. The experimental results show that, relative to CBA and CMAR, the classification accuracy and efficiency of CARPT are somewhat increased. The algorithm overcomes thedisadvantage of redundant nodes in CMAR algorithm and saves memory space significantly.Therefore, the algorithm is feasible and effective.
Keywords/Search Tags:Data Mining, Association Rule Mining, Classification Algorithm, Associative Classification, P-Trie tree
PDF Full Text Request
Related items