Research On Methods Of Data Mining Based On Granular Computing

Data Mining is a burgeoning technology which involves several disciplines. Data Mining can discover some latent and interesting knowledge that people don't known beforehand from the massive data, therefore it is also called Knowledge Discovery in the database. Data Mining have many research directions, and Association Rule Mining is the most important direction of them. Association Rule Mining can reveal the corelations between itemsets, and it can be widely applied to many fields such as market basket analysis, corelation analysis, classification, web-customised service.This thesis discusses a classical algorithm of Association Rule Mining- Apriori Algorithm and its transmutation. Aiming at some existed questions in Apriori Algorithm such as, seeking support of items need to scan databases many times and produceing candidata itemset need to match pattern, this thesis proposes one kind of Association Rule Mining algorithm based on Granular Computing---AR-GrC Algorithm. The AR-GrC Algorithm which imports granular computing thought regards each item as an information granular and obtains the binary expression of item through scanning one time of databases, then computes Support of itemset using AND and OR operation . The merit of AR-GrC Algorithm is that scanning database only once time and calculating Support of itemset by Granular Computing, which can reduce time-consuming and memory-taking, accordingly improve the efficiency of association rule mining.However, for many application, because the data is sparse in the multi-dimentional space, it is difficult to find strong association rules between original or underlying itemset. And then, this thesis proposes one kind of multi-level association rule mining algorithm based on the granular computing---ML-GrC Algorithm. The ML-GrC Algorithm mines association rule using AR-GrC algorithm on each level in multi-level structure and calculates Support of the itemset on high level applying hiberarchy of granular. It can discover strong association rules onany level, and improve the holistic mining efficiency through enhanceing mining efficiency on each level.