On Effective Association Rule Mining Algorithms for Temporal, Continuous and Big Data

• Main Authors: Kuo Cheng Yin, 尹國正
• Other Authors: Don Lin Yang
• Format: Others
• Language: en_US
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/99483196922371054393

博士 === 逢甲大學 === 資訊工程學系 === 102 === Data mining technology has been widely applied in various fields, while the association rule mining algorithm is the most representative technique. Our researches of association rule mining are divided into three major areas. The first of which is the rationality of discovered rules that are mined by association rule mining algorithm. The second is the adaptation of association rule mining algorithm when it is applied in different fields. The third is the mining efficiency improvement of association rule mining algorithm. In this dissertation, we identify potential problems in the three areas and propose effective algorithms to solve them respectively. In the study of the rationality of discovered rules, we present the concept of local frequent-itemsets, enhance the practicality of discovered association rules and develop a new algorithm to find out the global and local frequent-itemsets. In the study of the adaptation of association rule mining algorithm, we adjust traditional FP tree structure and develop a new algorithm to generate the approximate frequent-itemsets in different sliding windows under the environment of data stream. In the study of the mining efficiency improvement, we first propose a parallel processing algorithm to improve the mining efficiency under a cluster architecture. Second, we propose a compression algorithm to compress multiple transactions into a transaction. The purpose is to reduce the number of transactions to be mined, carry out the mining process in the main memory, and enhance the mining efficiency. In these studies, we performed various experiments to validate the proposed solutions and achieved our goals.