Font Size: a A A

DESIGN And IMPLEMENTATION Of GPU-based Association Rules Mining Algorithm

Posted on:2016-09-15Degree:MasterType:Thesis
Country:ChinaCandidate:X ZhangFull Text:PDF
GTID:2298330467992865Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
In recent years, with the rapid development of the Internet and computer technology, the ability of of data production,acquisition and storage using new technology has a big improvement, there has been more information stored in many fields with various froms. A lot of useful information is often hidden in these data. How to dig out this information is becoming a serious problem. Data Mining is the technology to solve the problem, the main goal is to find a set of database objects associated with some interesting or related links.Association rule mining is an important aspect of data mining technology which is focus on the relationship of different items in stead of the meaning of the items and order in database, and it has become a hot research topic and its results are widely used in marketing, business decisions and administrative management. It is a popular and well researched method for discovering interesting relations between variables in large databases and helps make better decision of strategy.A GPU-based parallel association rule mining algorithm is introduced in this paper, using the high parallel computing capability of GPU. This paper makes a deep research on GPU and CUDA programming architecture, and the classic algorithms of association rule mining and some well known improved versions. To find out a applicable way to design the parallel algorithm, this paper studys the detail procedure of association rule mining algorithm, and locate two mainly time consuming procedures, support counting and candidate sets generation procedure.With the cooperation of GPU and CPU by putting the logical operations in CPU and the complex parallel compution in GPU, this algorithm improves the efficiency of the supporting counting procedure with parallel processing, and improves the efficiency of the candidate sets generation procedure with the introduction of trie-based data starage structure. This paper chooses4testing data sets to run the algorithm and3other algorthms to do the performance comparison. Compared with other association rules mining algorithms, GPU based algorithm has a significant improvement in performance because of GPU’s high degree of parallelism and high memory bandwidth.
Keywords/Search Tags:association rule mining, GPU, CUDA, parallelcomputing
PDF Full Text Request
Related items