Airlines: Subsampled Tabular Benchmark with 10 Classes and 100 Features

A 2000-row, 100-column subsample of the airlines dataset from OpenML, created with a specific random seed. The subset contains up to 10 target classes and was generated using a stratified sampling method. The original dataset is in the public domain and was subsampled by Eddie Bergman.

Use Cases

• Benchmarking classification model performance based on the 10 target classes.
• Testing feature selection methods on the 100 sampled columns.
• Evaluating the impact of stratified sampling on model training for imbalanced datasets.

Strengths

• Subsampling parameters are explicitly defined: seed=1, nrows=2000, ncols=100, nclasses=10.
• Stratified sampling was applied to maintain class distribution in the subset.
• The dataset has a clear provenance as a derivative of the established OpenML 'airlines' dataset (ID 1169).

Limitations

• Row count is unknown for the original source data, limiting assessment of the subsample's representativeness.
• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

OpenML dataset 'airlines' (ID 1169)

Collection Method

Programmatic subsampling via a Python script using random and stratified selection.