Analysis on probabilistic and binary datasets through frequent itemset mining

Association rule mining is the process of discovering relationships among the data items in large database. It is one of the most important problems in the field of data mining. Finding frequent itemsets is one of the most computationally expensive tasks in association rule mining. The classical frequent itemset mining approaches mine the frequent itemsets from the database where presence of an item in a transaction is certain. Frequent itemset mining under uncertain data model is a new area of research. In this case the presence of an item is given by some likelihood measure. In this paper, we have developed a hyper structure based pattern growth method for frequent itemset mining from uncertain data. We have also developed a maximal clique based candidate pruning method for uncertain data. We have implemented and analyzed the performance of the well known algorithms for frequent itemset mining for both binary and uncertain data model. Our empirical results show that in case of dense binary datasets, FP-growth outperforms all other algorithms, whereas in case of sparse data H-mine outperforms other algorithms.