Pathfinder: A Bayesian Network for Medical Diagnosis with 109 Nodes

Pathfinder is a Bayesian network for medical diagnosis, originally developed by D. Heckerman, E. Horwitz, and B. Nathwani. The network contains 109 nodes, 195 arcs, and 72,079 parameters, with an average Markov blanket size of 3.82. The foundational work was published in Methods of Information in Medicine in 1992.

Use Cases

• Benchmarking structure learning algorithms based on the known network topology of 109 nodes and 195 arcs.
• Evaluating probabilistic inference methods based on the network's 72,079 parameters.
• Studying expert system design for medical diagnosis based on the network's original purpose.
• Analyzing the complexity of graphical models based on metrics like average degree (3.58) and maximum in-degree (5).

Strengths

• Well-defined network structure with 109 nodes and 195 arcs, providing a concrete benchmark.
• Detailed network statistics are provided, including 72,079 parameters and an average Markov blanket size of 3.82.
• Clear academic provenance with a cited 1992 publication by Heckerman, Horwitz, and Nathwani.

Limitations

• Column-level documentation is absent; variable semantics must be inferred from the original research.
• Row count and sample data are unavailable, which limits suitability assessment for specific modeling tasks.
• Last update date is unknown; freshness unverified.

Provenance

Source

D. Heckerman, E. Horwitz, and B. Nathwani, via the bnlearn Bayesian Network Repository.

Collection Method

Constructed as part of the Pathfinder expert system project for medical diagnosis.

Time Range

null

Freshness

unknown

Geography

null