Database Support for Data Mining Applications

This work presents a synthetic view on important concepts that have been studied within the cInQ European project when considering the pattern domain of itemsets, and introduces the concepts of pattern domain, evaluation functions, primitive constraints, inductive queries and solvers for itemsets.

Abstract

Mining Frequent Itemsets is the core operation of many data mining algorithms. This operation however, is very data intensive and sometimes produces a prohibitively large output. In this paper we give a complete set of rules for deducing tight bounds on the support of an itemset if the supports of all its subsets are known. Based on the derived bounds [l,u] on the support of a candidate itemset I, we can decide not to access the database to count the support of I if l is larger than the support threshold (I will certainly be frequent), or if u is below the threshold (I will certainly fail the frequency test). We can also use the deduction rules to reduce the size of an adequate representation of the collection of frequent sets; all itemsets I with bounds [l,u], where l =u, do not need to be stored explicitly. To assess the usability in practice, we implemented the deduction rules and we present experiments on real-life data sets.