Bayesian network decision models: structure, examples and how to build one
Learn how a Bayesian network decision model works, what its nodes and tables mean, and how to build one step by step for a clear, needs-based shortlist.
Contents
- Structure of the network
- A decision example in words
- Steps to build one
- Questions about Bayesian decision models
- How does a Bayesian network support decision making?
- What is the difference between a Bayesian network and an influence diagram?
- What data do I need to build a Bayesian network decision model?
- Can a Bayesian network handle several criteria at once?
A Bayesian network decision model connects uncertain facts to a choice, so you can see how evidence changes which option looks reasonable. The model shows variables as nodes linked by arrows, adds decision and utility nodes for the choice itself, and updates probabilities as new information arrives. This page explains the structure, walks through a decision example, and lists the steps to build one.
Structure of the network
A Bayesian network is a probabilistic graphical model built on a directed acyclic graph, meaning arrows point one way and no loop returns to its start. Each node holds a variable such as a cost driver or an outcome, and arrows mark which variables influence which. Conditional probability tables give the probability of each state of a node for every combination of its parents. Influence diagrams extend the graph with decision nodes, which are choices you control, and utility nodes, which score the outcomes.
A decision example in words
A Bayesian network decision example starts with the uncertain factors you cannot control and the choice you can. Suppose you compare two equipment options: the graph links expected workload and maintenance risk to a running cost, and a decision node holds the two options. A utility node scores each option against your priorities, such as low running cost or low downtime. When you enter evidence, such as a measured workload, evidence propagation updates the probability of each outcome and the expected utility of each choice.
Steps to build one
Building a Bayesian network decision model begins with listing the decision, the uncertain variables and the outcome you value. Draw the directed acyclic graph so each arrow reflects a real influence, then add decision and utility nodes for the choice. Fill in conditional probability tables from your own records and state clearly where a value is an estimate. Check the model by entering known cases and confirming the outputs match what happened, and involve a trained specialist when the graph or the math grows complex.
Questions about Bayesian decision models
How does a Bayesian network support decision making?
A Bayesian network supports decision making by showing how each piece of evidence shifts the probability of the outcomes you care about. You enter what you know, and evidence propagation recalculates the probabilities across the graph. Adding decision and utility nodes turns those probabilities into an expected score for each option, so you compare choices on the same footing.
What is the difference between a Bayesian network and an influence diagram?
An influence diagram is a Bayesian network extended with decision nodes and utility nodes for a choice. A plain Bayesian network describes uncertain variables and their conditional probability tables but has no place for options or scores. The influence diagram keeps the same directed acyclic graph and adds the decision and the value of each outcome.
What data do I need to build a Bayesian network decision model?
A Bayesian network decision model needs the variables, the arrows between them and a conditional probability table for each node. You can fill those tables from your own records, from expert judgment or from a mix, and you should label which is which. Where a value rests on judgment, treat it as an estimate and test the model against known cases.
Can a Bayesian network handle several criteria at once?
Yes, a Bayesian network can carry several criteria at once by giving each one a node and linking them to the outcome. A utility node then combines the criteria into one score that reflects the weights you set. This keeps the details of each option side by side, so you can see which criterion drives the result.