Computational complexity of itemset frequency satisfiability

Computing frequent itemsets is one of the most prominent problems in data mining. We introduce a new, related problem, called FREQSAT: given some itemset-interval pairs, does there exist a database such that for every pair the frequency of the itemset falls in the interval? It is shown in this paper that FREQSAT is not finitely axiomatizable and that it is NP-complete. We also study cases in which other characteristics of the database are given as well. These characteristics can complicate FREQSAT even more. For example, when the maximal number of duplicates of a transaction is known, FREQSAT becomes PP-hard. We describe applications of FREQSAT in frequent itemset mining algorithms and privacy in data mining.