Research On Association Rules Algorithm In Data Mining

At present, being a very essential research topic in data mining, association rules mining has achieved noticeable achievements, but in practical application, the efficiency of some association rules mining algorithms rapidly decreases with the increasing of data size. Therefore, these algorithms should be researched and improved in order to make them be applied more effectively and widely. Based on above,the dissertation mainly studies the algorithm of association rules mining, and presents two adaptive and effective association rules algorithms based on some existing algorithms.Firstly, the dissertation researches typical algorithms and their optimized algorithms of association rules. To overcome problems which exist in efficiency of candidate itemsets pruning and execution time of these algorithms, an improved and faster mining algorithm based on boolean matrix for short ABBM is proposed. The Boolean vector"relational calculus"method and association rules theorems are adopted in ABBM algorithm. Database is only scanned once and candidate itemsets aren't produced in mining process of ABBM algorithm, therefore, the times of producing frequent itemsets is decreased, consequently, execution performance and efficiency is heightened and performance optimization is achieved.Secondly, the dissertation researches typical incremental updating algorithms and their optimized algorithms. To resolves the problem of updating the association rules when increasing the database without changing the minimum support and minimum confidence, a high-efficient incremental updating algorithm for short HIUP is presented, HIUP discovers frequent itemsets of new database by AprioriTidList algorithm, and then classifies and prunes candidate itemsets in effective ways. So the times of scanning new database is decreased, and the efficiency of updating association rules is improved.To validate the capability of ABBM and HIUP algorithm, experiments are pursued on both synthetic and real databases. Experimental results show ABBM and HIUP algorithm have excellent performance in comparison with former algorithms, and the efficiency of algorithms rapidly increases with the increasing of data size. Therefore, these algorithms have preferable expansibility and wider application.