Mining rare itemsets using closed frequent itemsets with multiple item support thresholds
Most of mining methods use a single threshold to extract the whole set of frequent patterns. However, this assumption is not hold in real word applications since it does not reflect the nature of each item. In case the single minimum support threshold is set too low, a huge amount of itemsets will be generated including lots of redundant patterns. To avoid this problem, the single threshold should be set too high. But this cause a problem so-called rare itemsets since many interesting patterns may be lost. To tackle the rare itemset problem, lots of efforts has been studied to mine frequent patterns including rare ones. Recently, different Minimum Item Support thresholds (MIS) was considered instead of using single support threshold to generate complete set of frequent patterns without creating uninteresting patterns and losing substantial patterns. However, these methods are used to generate the complete set of frequent patterns including rare itemsets. Generating all frequent pattern including rare once is very expensive in term of time and memory as well. The main goal of this proposal is to improve an efficient method by which we can avoid generating redundant itemsets and useless patterns by utilizing the frequent closed itemsets mining with MIS framework. Since unknown knowledge (rare itemsets) is more interesting to the users, we extend this method to mine only most interesting itemsets (rare itemsets).