Study Of Compensated Fast Distributed Mining Algorithm Of Association Rules

Data mining acquires knowledge and rules that are implicit, unknown and having potential value for decision-making from large databases or data warehouses. It is the result that combines artificial intelligence and database. At present, data mining is one of the most advanced research direction in the field of database and information decision.Association rules mining is active data mining research area and applies more widely than other methods. In the dissertation, it introduces the basic concepts, characteristics and famous algorithms of association rules mining in detail. Existing algorithms and modules cater to a centralized environment, such as database or data warehouse. With the development of distributed database and network technology, they do not meet the needs of mining rules from distributed data sets. The interest in distributed association rules mining arises from this situation.At present, most of the distributed or paralleling mining algorithms are developed from corresponding sequential ones. So the basic concepts, characteristics and famous sequential AR algorithms are introduced in the dissertation at first. Then FDM algorithm and the techniques that are used with FDM algorithm are referred. FDM algorithm uses the relationships between the locally large and globally large itemsets to generate a smaller number of globally candidate sets. The local pruning and global pruning are finished at each site. To reduce communication costs, a count polling technique is introduced to compute the support counts of candidate sets。Through the analysis of FDM algorithm, we find that FDM algorithm will lose some frequent k-itemsets when it is used to compute all frequent k-itemsets on distributed database D. Because of this, the result computed by FDM algorithm will be different from the real one. To be aimed at the shortage of FDM algorithm, the dissertation advances CFDM algorithm. It succeeds in solution of losing frequent k-itemsets question with FDM algorithm. With CFDM algorithm, we can get...