CoIL 2000: Insurance Policy Risk Classification Data

A 2000 dataset from the CoIL Challenge, curated by the TabArena team for evaluating predictive models on independent and identically distributed tabular data. It contains anonymized insurance policy records for a binary classification task. The original data was sourced from a technical report by the Leiden Institute of Advanced Computer Science.

Use Cases

• Benchmarking binary classification algorithms based on the described insurance policy features.
• Studying model performance on IID tabular data as part of the TabArena study framework.
• Developing risk prediction models for insurance applications based on the dataset's intended task.

Strengths

• Features have been curated with semantic, meaningful names.
• Original training and validation data were combined into a single dataset.
• Ordinal encoding from the original data was reversed where possible.

Limitations

• The dataset contains 15% duplicate records, which is noted as an anomaly.
• Column-level documentation, sample data, row count, and file formats are unknown.
• The last update date is unknown, making freshness unverified.

Provenance

Source

Technical Report 200009, Leiden Institute of Advanced Computer Science, Universiteit van Leiden.

Collection Method

Curated from the original CoIL challenge data source by the TabArena team.

Time Range

2000

Freshness

Dataset year is 2000; last updated date is unknown.

Geography

null