Hayes-Roth: A Synthetic Classification Benchmark Dataset

A merged version of the separate train and test sets for the Hayes-Roth database, originally created by Barbara and Frederick Hayes-Roth in March 1989. The dataset contains synthetic instances with five nominal attributes, including hobby, age, education level, and marital status, and a class label determined by a specific rule. It is designed as a classification benchmark where only three of the five attributes are diagnostic.

Use Cases

• Testing classification algorithm performance on a synthetic, rule-based dataset.
• Demonstrating feature selection concepts, as only three of the five attributes are described as diagnostic.
• Teaching classification principles using a dataset with a clearly defined decision rule.
• Benchmarking model interpretability on data with known, simple class prototypes.

Strengths

• Clear, rule-based class definitions are provided in the description, aiding interpretability.
• Dataset origin and creators are explicitly documented.
• The dataset is a merged version of standard train/test splits, providing a complete set for analysis.

Limitations

• Row count, column details, and file formats are unknown, limiting suitability assessment.
• Data is synthetic and generated according to a specific 1989 experiment, which may limit real-world relevance.
• Column-level documentation beyond the listed attribute names is absent; field semantics must be inferred.

Provenance

Source

Barbara and Frederick Hayes-Roth, donated by David W. Aha.

Collection Method

Synthetically generated for a classification experiment.

Time Range

Data created in March 1989.

Freshness

Last update date is unknown; freshness unverified.

Geography

Synthetic data; no geographic coverage.