Distributed implementation of an intelligent data classifier

Abstract

Industry, science and business applications need to manipulate a huge amount of data every day. Most of the time these data come from distributed sources and are analyzed trying to discover knowledge and recognize patterns using Data Mining techniques. Data classification is a technique that allows to decide if a set of data belongs to a group of information or not. Data classification requires putting all data together in a big centralized datasets. To congregate and analyze this dataset represents a very expensive task in terms of time, memory and bandwidth consuming. Nowadays, architectures for Distributed Data Mining have been developed trying to reduce computing and storage costs. This paper presents an approach to building a distributed data classifier which takes only metadata from distributed datasets avoiding the total access to the original data. Using only metadata reduces the computing time and bandwidth consumption required to build a data classifier. © 2010 Springer-Verlag Berlin Heidelberg.