A Study on mining association rules with Incremental updates

• Main Authors: Chung-Yung Wu, 吳忠勇
• Other Authors: Don-Lin Yang
• Format: Others
• Language: zh-TW
• Published: 2002
• Online Access: http://ndltd.ncl.edu.tw/handle/pw465k

碩士 === 逢甲大學 === 資訊工程所 === 90 === Although many efficient algorithms have been proposed for the discovery of association rules and the greater part of these algorithms can obtain good performance. But there is a serious problem in these algorithms. As long as the original database changed and then the only choice is to rerun the static algorithm of association rules mining once. Unfortunately the database will have a high probability to vary in the practical application. The problem that we call incremental mining of association rules is focusing on how to quickly update the association rules of the updated database and it is interesting for many people. In this thesis, in order to efficiently update the frequent itemsets of the updated database, we presented two efficient algorithms to handle the incremental maintenance of association rules. For convenience we call the two algorithms EIM-A and EIM-G simply. EIM-A is focusing on only there are new data added in the original database. It only needs to scan the original database less then once and using the frequent itemsets that contain in the knowledge database to create filter condition. By the condition filter the candidate itemsets that large in incremental data. The best case is never to scan the original database. The worst case is just using the large itemset generated from incremental data to rescan the original database. EIM-G not only handle add data but also data is deleted from original database. EIM-G utilizes the minimal infrequent itemset that contained in the knowledge database to generate new candidate itemsets. EIM-G also maintain the hashing table that in the knowledge database in order to avoid the number of candidate itemsets to be too huge. The experiment results will prove the peroformance of EIM-G. EIM-G will have a good performance especially when the number of original database is much large then the number of incremental data.