Pathfinder 4: A Bayesian Network for Medical Diagnosis

109 nodes and 195 arcs define the Pathfinder Bayesian Network, a probabilistic model for lymph-node pathology diagnosis. The network contains 72,079 parameters and was authored by D. Heckerman, E. Horwitz, and B. Nathwani. It serves as a foundational benchmark for normative expert systems research.

Use Cases

• Benchmarking Bayesian network structure learning algorithms based on the known graph topology.
• Evaluating probabilistic inference algorithms based on the network's 72,079 parameters.
• Studying the design of normative expert systems for medical diagnosis, as referenced in the original Pathfinder project.
• Analyzing the properties of large discrete networks based on metrics like average Markov blanket size (3.82) and maximum in-degree (5).

Strengths

• Well-defined structure with 109 nodes and 195 arcs, providing a concrete benchmark.
• Large parameter space of 72,079 parameters for complex probabilistic modeling.
• Clear academic provenance with a canonical citation to the 1992 Pathfinder Project paper.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for certain machine learning tasks.
• Last update date is unknown; freshness unverified.

Provenance

Source

bnlearn Bayesian Network Repository, referencing the Pathfinder Project.

Collection Method

Constructed as a normative expert system for lymph-node pathology diagnosis.

Time Range

null

Freshness

null

Geography

null