Minimum Description Length Adaptive Bayesian Mining

View Sample PDF

Author(s): Diego Liberati (Italian National Research Council, Italy)
Copyright: 2009
Pages: 5
Source title: Encyclopedia of Data Warehousing and Mining, Second Edition
Source Author(s)/Editor(s): John Wang (Montclair State University, USA)
DOI: 10.4018/978-1-60566-010-3.ch191

Purchase

View Minimum Description Length Adaptive Bayesian Mining on the publisher's website for pricing and purchasing information.

Abstract

In everyday life, it often turns out that one has to face a huge amount of data, often not completely homogeneous, often without an immediate grasp of an underlying simple structure. Many records, each instantiating many variables are usually collected with the help of several tools. Given the opportunity to have so many records on several possible variables, one of the typical goals one has in mind is to classify subjects on the basis of a hopefully reduced meaningful subset of the measured variables. The complexity of the problem makes it natural to resort to automatic classification procedures (Duda and Hart, 1973) (Hand et al., 2001). Then, a further questions could arise, like trying to infer a synthetic mathematical and/or logical model, able to capture the most important relations between the most influencing variables, while pruning (O’Connel 1974) the not relevant ones. Such interrelated aspects will be the focus of the present contribution. In the First Edition of this encyclopedia we already introduced three techniques dealing with such problems in a pair of articles (Liberati, 2005) (Liberati et al., 2005). Their rationale is briefly recalled in the following background section in order to introduce the kind of problems also faced by the different approach described in the present article, which will instead resort to the Adaptive Bayesian Networks implemented by Yarmus (2003) on a commercial wide spread data base tool like Oracle. Focus of the present article will thus be the use of Adaptive Bayesian Networks are in order to unsupervisedly learn a classifier direcly form data, whose minimal set of features is derived through the classical Minimun Description Lenght (Barron and Rissanen, 1998) popular in information theory. Reference will be again made to the same popular micro-arrays data set also used in (Liberati et al., 2005), not just to have a common benchmark useful to compare results and discuss complementary advantages of the various procedures, but also because of the increasing relevance of the bioinformatics field itself.

The IRMA Community

Research IRM

Minimum Description Length Adaptive Bayesian Mining

Purchase

Abstract

Related Content

IRMA Sponsors