Frequent closed itemset based algorithms

Abstract

As a side effect of the digitalization of unprecedented amount of data, traditional retrieval tools proved to be unable to extract hidden and valuable knowledge. Data Mining, with a clear promise to provide adequate tools and/or techniques to do so, is the discovery of hidden information that can be retrieved from datasets. In this paper, we present a structural and analytical survey of <u>f</u>requent <u>c</u>losed <u>i</u>temset (FCI) based algorithms for mining association rules. Indeed, we provide a structural classification, in four categories, and a comparison of these algorithms based on criteria that we introduce. We also present an analytical comparison of FCI-based algorithms using benchmark dense and sparse datasets as well as "worst case" datasets. Aiming to stand beyond classical performance analysis, we intend to provide a focal point on performance analysis based on memory consumption and advantages and/or limitations of optimization strategies, used in the FCI-based algorithms.