tabullm.load_fraud

tabullm.load_fraud(data_dir=None, return_metadata=True)[source]

Load and preprocess the Real or Fake Job Posting Prediction dataset.

If the dataset is not found locally, downloads it automatically from Zenodo (no credentials required). If the download fails, a FileNotFoundError is raised with manual download instructions.

Parameters:

  • data_dir (str or Path, optional) – Directory containing fake_job_postings.csv (or where it will be downloaded). Defaults to ~/.tabullm/fraud/.

  • return_metadata (bool, default=True) – If True, returns (X, y, metadata). If False, returns (X, y).

Returns:

  • X (pandas.DataFrame, shape (n_samples, 15)) – Features: 7 text columns, 3 binary columns, 5 categorical columns.

  • y (pandas.Series, shape (n_samples,)) – Target variable (fraudulent: 0 = legitimate, 1 = fraudulent).

  • metadata (dict) – Dataset metadata including column categorization, class distribution, and missing value summary. Only returned when return_metadata=True.

Notes

The dataset is highly imbalanced (~4.84% fraud). Consider using class_weight="balanced" in downstream classifiers.

Dataset: Real or Fake Job Posting Prediction (Vidros et al., 2017) Original source: https://www.kaggle.com/datasets/shivamb/real-or-fake-fake-jobposting-prediction Zenodo archive (concept DOI): https://doi.org/10.5281/zenodo.18884001 License: ODbL v1.0