Frequent itemset mining from databases including one evidential attribute

Abstract

Frequent Itemset Mining (FIM) problem has been extensively tackled in the context of perfect data. However, real applications showed that data are often imperfect (incomplete and/or uncertain) which leads to the need of FIM algorithms that process imperfect databases. In this paper we propose a new algorithm for mining frequent itemsets from databases including exactly one evidential attribute. An evidential attribute is an attribute that could have uncertain values modelled via the evidence theory, i.e., a basic belief assignment. We introduce in this paper a variant of the structure Belief Itemset Tree (BIT) for mining frequent itemsets from evidential data and we lead some experiments that showed efficiency of our mining algorithm compared to the existing ones. © 2008 Springer-Verlag.