Algoritmos de mineração de dados eficiente quanto ao consumo de memória

Abstract

The volume of data input to data mining applications has grown considerably as an indirect consequence of the price reductions for data aquisition, transmission and storage. Thus, data mining applications must be scalable, that is, the losses in performance should be small when the size of the input is increased. Frequent itemset mining is a popular data mining application for which there are several algorithms and implementations. EClaT is among the most successful and wellknown algorithms. Its most memory consuming abstract data type is the natural number set. In this work, we replaced the implementation for this abstract data type for another, commonly employed by information retrieval algorithms but never before employed by data mining algorithms, that saves memory. We adapted to the new context and/or implemented other memory saving techniques as well. We achived an economy in maximum memory consumption of up to an order of magnitude compared to the original implementation.