Nine synthetic datasets designed to contain planted traps for common machine learning pitfalls. The collection likely contains examples of data leakage, bias, drift, and confounding variables. It is hosted on Kaggle, but specific details about the creator, size, and license are not provided.
Use Cases
- Testing model robustness against data leakage based on the planted leakage traps.
- Evaluating bias detection and mitigation algorithms based on the planted bias traps.
- Benchmarking drift detection methods based on the planted drift traps.
- Studying the effects of confounding variables on model performance based on the planted confounding traps.
Strengths
- Contains nine distinct datasets, each with a specific focus.
- Explicitly designed to include known pitfalls like leakage, bias, drift, and confounding for testing purposes.
Limitations
- Row count and dataset size are unknown, which may limit suitability assessment.
- 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
- Kaggle
- Collection Method
- Synthetically generated for testing purposes.