HTRU2: Pulsar Detection Dataset with 17,898 Candidates

17,898 candidate signals for identifying pulsars, a rare type of neutron star. The dataset contains 1,639 positive and 16,259 negative examples, each described by 8 continuous variables derived from integrated pulse profiles and DM-SNR curves. It originates from research by Dr. Robert Lyon at the University of Manchester and is hosted on OpenML.

Use Cases

• Train binary classifiers to distinguish pulsar signals from noise based on integrated profile statistics.
• Develop feature-based anomaly detection models using the DM-SNR curve variables.
• Benchmark classification algorithms on an imbalanced astronomical dataset.
• Study the statistical properties of radio signals for astrophysical research.

Strengths

• Contains 17,898 total examples with a clear class distribution of 1,639 positives and 16,259 negatives.
• Each candidate is described by 8 continuous variables derived from two distinct astrophysical measurements.
• The dataset has a specific scientific purpose for pulsar detection and is associated with documented research.

Limitations

• Column-level documentation is absent; field semantics must be inferred from the high-level description.
• Last update date is unknown; freshness unverified.
• The dataset is imbalanced, which may require specific modeling techniques.

Provenance

Source

University of Manchester, School of Physics and Astronomy (Dr. Robert Lyon). Also referenced on the UCI Machine Learning Repository.

Collection Method

Likely collected from radio telescope observations, with features extracted from integrated pulse profiles and DM-SNR curves.