Added description of freMTPL2 dataset to README.md

README.md

@@ -14,4 +14,72 @@ size_categories:
 - 100K<n<1M
 language:
 - en
----
+---
+
+# `freMTPL2` Dataset
+
+This dataset is a mirror of the freMTPL2 frequency and severity datasets, originally published by Arthur Charpentier to accompany his textbook [*Computational Actuarial Science with R*](https://www.routledge.com/Computational-Actuarial-Science-with-R/Charpentier/p/book/9781138033788).
+
+The freMTPL2 dataset contains data on Third-Party Liability (TPL) Motor insurance policies issued in France, along with claims filed against those policies, observed over a duration of just over a year.
+
+These observations are organized into two separate CSV files:
+1. `freMTPL2freq.csv`: a 'Frequency' dataset that stores the risk attributes of the observed insurance policies.
+2. `freMTPL2sev.csv`: a 'Severity' dataset that records monetary loss of each claim made against one of the policies in `freMTPL2freq.csv`.
+
+For further details on the `freMTPL2` dataset, please consult [Page 71 of the `CASdatasets` R package documentation](https://cas.uqam.ca/pub/web/CASdatasets-manual.pdf#page=71).
+
+Both `freMTPL2freq.csv` and `freMTPL2sev.csv` in this HuggingFace dataset were sourced from the [karansarpal/freMTPL2-french-motor-tpl-insurance-claims Kaggle dataset](https://www.kaggle.com/datasets/karansarpal/freMTPL2-french-motor-tpl-insurance-claims).
+
+## `freMTPL2freq.csv` Description
+
+This file contains data on 678,013 TPL motor insurance policies. Each row represents a single policy and contains the following columns:
+- `IDpol` (`int`): Policy ID number; uniquely identifies each row in `freMTPL2freq.csv`.
+- `ClaimNb` (`int`): Total number of TPL motor claims made against the policy.
+- `Exposure` (`float`): Duration (in years) the policy was observed.
+- `VehPower` (`int`): Vehicle power encoded as an ordinal integer (higher values indicate more powerful vehicles).
+- `VehAge` (`int`): Age of the vehicle in years (integer part only).
+- `DrivAge` (`int`): Driver's age in years (integer part only). In France, the legal driving age is 18. The minimum driver age in this dataset is 18 years old.
+- `BonusMalus` (`int`): [Bonus-Malus factor](https://en.wikipedia.org/wiki/Bonus%E2%80%93malus#Insurance) that ranges from 50 to 350. In France, < 100 means bonus, whilst > 100 means malus.
+- `VehBrand` (`string`): Anonymized vehicle brand code.
+- `VehGas` (`string`): Fuel type, either 'Regular' or 'Diesel'.
+- `Area` (`string`): Population density category of the city/community the driver lives in; these categories are ordered from "A" (rural) to "F" (urban center).
+- `Density` (`int`): Number of inhabitants per square-kilometer living in the city/area where the vehicle driver resides.
+- `Region` (`string`): Region in France the vehicle driver resides; these categories correspond to the [1982 - 2015 French region definitions](https://en.wikipedia.org/wiki/Regions_of_France#1982%E2%80%932015).
+
+## `freMTPL2sev.csv` Description
+
+This file contains data on 26,639 claims made against policies in `freMTPL2freq.csv`. Each row represents a single claim and contains the following columns:
+- `IDpol` (`int`): Policy ID number linking the claim to a policy in `freMTPL2freq.csv`. Note that `IDpol` is *not* a unique row identifier in this file, as multiple claims may be made against the same policy.
+- `ClaimAmount` (`float`): Most up-to-date estimate of the total financial loss/cost of the claim, presumably in Euros (although the exact currency the financial losses are recorded is not disclosed).
+
+## Relationship between `freMTPL2freq.csv` and `freMTPL2sev.csv`
+
+There is a one-to-many relationship between `freMTPL2freq.csv` and `freMTPL2sev.csv` such that:
+- Each claim in `freMTPL2sev.csv` is associated with no more than one policy in `freMTPL2freq.csv`.
+- Each policy in `freMTPL2freq.csv` can join onto one or more claims in `freMTPL2sev.csv` (i.e. multiple claims can be made against the same policy).
+- Conversely, not all policies in `freMTPL2freq.csv` will join onto a claim listed in `freMTPL2sev.csv` (i.e. not all policyholders make a claim against their policy). Indeed, ~96% of the policies in `freMTPL2freq.csv` have no associated claims in `freMTPL2sev.csv`.
+
+## Data Inconsistency Issues
+
+Please be aware that some data inconsistencies exist between the `freMTPL2freq.csv` and `freMTPL2sev.csv` files:
+- *Missing Policy Records*: Not all claims in `freMTPL2sev.csv` have a corresponding policy in `freMTPL2freq.csv`, meaning that the risk attributes associated with these claims are unknown. This issue affects 195 out of 26,639 records in `freMTPL2freq.csv`.
+- *`ClaimNb` Discrepancies*: For some policies in `freMTPL2freq.csv`, the recorded `ClaimNb` does not match the actual number of claims recorded in `freMTPL2sev.csv` for those policies. For instance:
+   - Some policies with `ClaimNb = 1` have no matching claims in `freMTPL2sev.csv`.
+   - One policy with `ClaimNb = 2` has only one matching claim in `freMTPL2sev.csv`.
+   
+  This issue affects 9,117 out of 678,013 policies in `freMTPL2freq.csv`.
+
+These data inconsistency issues have been retained in this Hugging Face dataset so that it remains an exact replica of the original freMTPL2 dataset.
+
+## Referencing
+
+If you use this dataset in your work, please cite the original publishers of the freMTPL2 dataset:
+
+```latex
+@article{dutang2020package,
+  title={Package ‘casdatasets’},
+  author={Dutang, Christophe and Charpentier, Arthur},
+  journal={url: https://dutangc.perso.math.cnrs.fr/RRepository/pub/web/CASdatasets-manual.pdf},
+  year={2020}
+}
+```