Enhanced approximation of the emerging pattern space using an incremental approach

From the many different patterns that can be extracted from data, so-called emerging patterns (EPs) are a particular useful kind. EPs are itemsets whose supports increase significantly from one dataset to another. Existing methods used to discover EPs have been successfully applied only under a constrained search space. Although they may provide a very efficient way of discovering some sort of EPs, they are rather limited when the whole set of EPs is needed, as they just compute an approximation of that set. Recent EPs techniques rely on borders, a concise representation of the candidate itemsets which does not require computing an exponentially large number of such candidates. In this paper we outline a new method which exploits previously mined data using an incremental approach, requiring thus less dataset accesses. Our proposal also aims to reduce the amount of work needed to perform difference operations among borders taking into account special properties of the itemsets. Eje: Inteligencia artificial Red de Universidades con Carreras en Informática (RedUNCI)