Bayesian network

(redirected from Bayes networks)

Bayesian network

A form of artificial intelligence—named for Bayes’ theorem—which calculates probability based on a group of related or influential signs. Once a Bayesian network AI is taught the symptoms and probable indicators of a particular disease, it can assess the probability of that disease based on the frequency or number of signs in a patient.
Mentioned in ?
References in periodicals archive ?
According to the results of the calculations, the variable describing the economy which in the period 2000-2013 was the most frequently reported in Bayes networks describing the relationship of the size of general government sector was the parameter that is called gross domestic product in current prices per inhabitant (GDP per inhabitant).
Subsequet positions were taken by the following ratios: FDI--foreign direct investment (million USD)--39 occurrences in networks, retail sales--dynamic index of turnover (total 2010 = 100)--37 occurrences, growth rates of GDP (percentage change)--36 occurrences, as well as inward FDI flows (million USD)--32 occurrences in Bayes networks and potential output of total economy (million euro)--30 occurrences.
There were seven ratios that gained five occurrences in Bayes networks. These were: inward FDI flows (million USD) in year 2012, potential output of total economy (million euro) in year 2000 and 2005, potential output of total economy (dynamic annual average rate of growth-percentage) in year 2010, as well as gross domestic product in current prices per inhabitant-dynamic (percentage change) in year 2004, harmonized indices of consumer prices (HICPs) (annual average rate of change) in year 2013, outward FDI flows (million USD) in year 2000 and unemployment rate (in %) in 2009.
First of all, the research provided a creation of a ranking of the maximum number of occurrences for variables that describe an economy in Bayes networks (describing a relation of an economy to the size of the general government sector) in the examined period 2000-2013 (see Table 5).
Some of these studies can be referred as Bayes networks [5], rough sets [6], and artificial neural networks [7] which are the examples of such studies.
Two algorithms have been applied to this Bayes network; one of them is a message-passing algorithm, developed by Pearl [55] to update the probabilities in the various linked networks using conditioning methods and the second one is that the exact inference algorithm, developed by Lauritzen and Spiegelhalter [56] for local probability calculations in the graphical structure.
The probabilities in the Bayes network can represent both objective and subjective information.
More complex interactions between predictors may be represented by augmented naive Bayes networks by allowing directed edges (probabilistic dependencies) between predictors, dispensing with the strong independence assumption of naive Bayes.
These include decision trees, influence diagrams, and Bayes networks. These quantitative techniques assist the analyst not only in describing the problem and communicating information about structure, but also in calculating the effect of the truth of one proposition or piece of evidence on the plausibility of others.
It is a snapshot of a problem at a particular instant in time and thus, as has been pointed out by many researchers, it fails to capture the dynamic, even messy, nature of real-life problems.(57) Bayes networks enable one to process information and to move from a description of the problem predata to a description of the problem postdata.
Despite this, the standard exposition of both Bayes networks and decision trees assumes a defined problem and a predetermined set of influences.
The most obvious difference between Wigmore Charts and Bayes networks is that no mechanism for assessing probabilities exists in Wigmore Charts.