Study On Mining Maximal Frequent Itemset Based On Iceberg Concept Lattice

In the process of association rule mining, the sovlving of frequent itemset is the basis and prerequisite for association rule mining, which is the most time-consuming step.Reduce the number of candidate items is the best way to reduce the computational cost.Since maximal frequent itemset has been implied all frequent itemset, the problem of finding frequent itemset can be converted into the problem of finding maximal frequent itemset. The size of itemsets can be reduced effectively by finding maximal frequent itemset, so user can find the knowledge from the dense data set quickly.However, with the quick development of data collection and data storage technique, many institutions and organizations have stored a large scale of dataset, which result in the low efficiency of the existing maximal frequent itemset mining algorithm. The chief problem is that scanning database frequently needs plenty of I/O cost; a large scale of candidate dataset leads to impossibility of calculation and storing them; contributing to the difficulty of incremental mining; it will have a lot of redundant and uninterested rules and so on. To solve the above problems, our article put the iceberg concept lattice modle to the study of maximal frequent itemset mining.Iceberg concept lattice is partial order lattice structure composited by all frequent concepts from concept lattice under the condition of a user-specified support threshold. Each intent of frequent concept is a closed frequent itemsets, which describes relationships of the data set objects and attributes. Using relationship of frequent closed itemset and maximal frequent itemset and good generalization and specialization bewtween frequent concept.Incremental maximal frequent itemsets mining algorithms based on the maximum frequent itemset and the growth properties is proposed.Our paper also research the positive non-redundant association rules based on iceberg concept lattices, Solved association rules mining from maxinal frequent itemset will lead to the loss of support, and there is a lot of information the user is not interested in redundant rules.The main contribution of the paper is as follows:(1) Propose the alogithm of mining Maxiaml Frequent Itemset based on iceberg concept lattice ICMFIA (Iceberg Concept Lattice Maximal Frequent Itemset Alogithm). The algorithm builts the iceberg concept lattice by scanning the data sets at a time, using the coverage relationship between frequent concepts in the iceberg concept lattice, we can quickly calculate the maximum frequent concepts corresponding to the maximum frequent itemsets. The intension of all maximal frequent concepts are the set of all maximal frequent itemsets.The theory and experimental results show that the proposed ICMFIA algorithm outperforms the other existing algorithm in a fewer number of scanning data sets and Mining efficiency.(2) Propose the algorithm of incremental mining maximal requent itemset MFI-AI (Maximal Frequent Itemset-Attribute Incremental) based on iceberg concept lattice. The algorithm mainly mining maximal frequent itemset after increaseing the attributes of dataset.On the basic of the original iceberg concept lattices,it construct a new iceberg concept lattices and calculate the updating and new Maximal Frequent concept.Then, just mine the new maximal frequent itemsets and need to updating maxiaml frequent itemsets in the set of maximal frequent itemset. The algorithm avoid re-mine all the maximal frequent itemset after increaseing the attributes of dataset. The experimental results show our algorithm with the merit of less repetitive tasks and efficiency(3) Propose the method of mining positive relevant and non-redundant association rules in iceberg concept lattice solved the information loss of support and a large number of uninterested association rules, which mine from maximal frequent. The scale of mining rules is reduced by non-redundant Association Rules. At the same time, support and confident of other functional association rules can be calculated by Non-Redundant Association Rules. Since calculation of strong rules is not interesting while using support-confident framework, Lift for mining positive non-redundant association rules is introduced, mining the rules which user really interested.