Information Technology Reference
In-Depth Information
Chapter 6
Probabilistic Reasoning
6.1 Introduction
Bayesian Network is a graphic model for describing the connection probabilities
among variables. It provides a natural representation of casual relationship and is
often used to explore potential relationships among data. In the network, nodes
represent variables and directed links represent dependant relationships between
variables. With firm mathematical foundation, Bayesian Theory offers method
for brief function calculation, and describes the coincidence of brief and evidence,
and possesses the incremental learning property that brief varies along with the
variation of evidence. In data mining, Bayesian networks can deal with
incomplete or noisy data set. It describes correlations among data with
probabilistic measurement, and thereby solves the problem of data inconsistency.
It describes correlations among data with graphical method, which has clear
semantic meaning and understandable representation. It also makes prediction
and analysis with casual relationships among data. Bayesian network is
becoming one of the most remarkable data mining methods due to its nice
properties, including unique knowledge representation of uncertain information,
capability for handling probability, and incremental learning with prior
knowledge.
6.1.1
History of Bayesian theory
The foundational work of Bayesian School is Reverend Thomas Bayes'
(1702-1761) “An Essay towards solving a Problem in the Doctrine of Chances”.
Maybe he felt the work was not perfect enough, this work was published not in
his lifetime, but posthumously by his friend. As famous mathematician Laplace
P. S. educed Law of Succession based on Bayesian method, Bayesian method
and theory began to be recognized. In 19th century, because the problem of
214
