Bayesian Networks for Expert Systems: Theory and Practical Applications

Abstract

Bayesian network are widely accepted as models for reasoning with uncertainty. In this chapter we focus on models that are created using domain expertise only. After a short review of Bayesian networks models and common Bayesian network modeling approaches, we will discuss in more detail three applications of Bayesian networks. With these applications, we aim to illustrate the modeling power and flexibility of the Bayesian networks that goes beyond the standard textbook applications. The first network is applied in a system for medical diagnostic decision support. A distinguishing feature of this network is the large amount of variables in the model. The second one involves an application for petrophysical decision support to determine the mineral content of a well based on borehole measurements. This model differs from standard Bayesian networks by its continuous variables and nonlinear relations. Finally, we will discuss an application for victim identification by kinship analysis based on DNA profiles. The distinguishing feature in this application is that Bayesian networks are generated and computed on-the-fly based on case information.

What to measure next to improve decision making? On top-down task driven feature saliency

“…We further tested the single feature information content as shown in figure 3 panel (e) and find that all features are informative (p < 0.01), with feature 8 as the most informative with almost 0.4 bits of information. We choose in this case to provide features (1,2) and evaluate the top down saliency for features (3,4,5,6,7,8).…”

“…We simulate incomplete test measurement, in which are given the two features (1,5). The panels show: (a) Error rates in the training set of complete data (N train = 650) and in the test set (Ntest = 242) using complete data, test error rate using only the initial feature set (1,5), and test error using (1,5), and the feature chosen by the top down saliency estimate, and finally the test error obtained using (1, 5) and a randomly chosen additional feature; (b) Estimated information saliency obtained on the test data, given the incomplete feature vector (1, 5); (c) Frequency of selection of the additional features; (d) Frequency of selection of features in test cases within the two classes; (e) The log 2 mutual information between features and class label. noisy decision problem than the previous Abalone case.…”

“…Each data sample is based on the record of a single male individual [10]. We tested the single feature information content as shown in figure 2 panel (e) and find that only features (5,6) are informative on their own (p < 0.01). This data set is smaller with N train = 200 and N test = 95.…”

“…While not as well studied as the basic missing data problem, the sequential measurement problem has been studied in several application areas including active vision [3], [4], petrophysics and medical diagnosis [5], geophysics [6], and robot navigation [7]. These previous efforts all use an information theoretic approach but implemented in specialized domain dependent representations, hence, less general than the present approach.…”

An Emergency Unit Support System to Diagnose Chronic Heart Failure Embedded with SWRL and Bayesian Network

“…The graphical representation that shows the conditional independencies between the nodes are easy to be understand by the user of the system. Moreover, since BN defines a unique joint between two specific nodes, the consistency and correctness of inference are guaranteed due to the mathematical calculation dependencies [4].…”