Insurance/description.md

Data Challenge Overview

Welcome to the data challenge! This dataset release contains information for three predictive modeling tasks involving auto insurance policies and claims.

Directory & File Structure

The dataset is organized into CSV files and two folders containing images and PDF documents:

├── train_claims.csv
├── test_claims.csv
├── train_policies_subset.csv
├── test_policies_subset.csv
├── vehicle_images/
│   ├── train/
│   │   ├── <PolicyID>.jpg
│   └── test/
│       ├── <PolicyID>.jpg
├── invoices/
│   ├── train/
│   │   ├── invoice_<ClaimID>.pdf
│   └── test/
│       ├── invoice_<ClaimID>.pdf

Claims Tables

Files: train_claims.csv, test_claims.csv

Columns:

• ClaimID: Unique identifier for each claim.
• PolicyID: Identifier linking the claim to a policy.
• ClaimDate: Date the claim occurred.
• ClaimType: Type of claim ("Fender-Bender", "Major Collision", "Theft/Comprehensive").
• ReportedDamage: Reported monetary damage of the claim.
• NumParties: Number of parties involved in the claim.
• Description: Textual description of the claim.
• ClaimComplexityLabel (train only): Label indicating claim complexity ("Simple", "Moderate", "Complex").
• FraudLabel (train only): Binary indicator (0 or 1) of whether the claim is fraudulent.

Note: ClaimComplexityLabel and FraudLabel are not provided in the test set.

Policies Tables

These policy tables are a subset of policies—only those for which a vehicle image is available.

Files: train_policies_subset.csv, test_policies_subset.csv

Columns:

• PolicyID: Unique identifier for the insurance policy.
• HolderAge: Age of the policyholder.
• VehicleType: Type of vehicle (Sedan, SUV, Sports).
• AnnualMileage: Miles driven annually by the vehicle.
• LocationUrban: 1 if urban, 0 if rural.
• CreditScore: Credit score of policyholder (0 to 1 scale).
• PolicyStart: Policy start date.
• PolicyEnd: Policy end date.
• NextYearLoss (train only): Total insurance loss expected for the next year (target variable).

Note: NextYearLoss is not provided in the test set.

Image & PDF Folders

vehicle_images/train/ and vehicle_images/test/:

Contains images named as <PolicyID>.jpg, visually indicating the vehicle's condition.

invoices/train/ and invoices/test/:

Contains PDF invoices for each claim, named as invoice_<ClaimID>.pdf.

Challenge Tasks & Submission Format

Challenge 1: Claims Complexity Prediction

Predict the ClaimComplexityLabel for each claim in test_claims.csv.

Submission CSV: ClaimID,ClaimComplexityLabel

Evaluation Metric: Macro-F1.

Challenge 2: Risk-Based Pricing

Predict the NextYearLoss for each policy in test_policies_subset.csv.

Submission CSV: PolicyID,NextYearLoss

Evaluation Metric: Normalized Gini.

Challenge 3: Fraud Detection

Predict the binary FraudLabel for each claim in test_claims.csv.

Submission CSV: ClaimID,FraudLabel

Evaluation Metric: Macro-F1.

Notes & Tips

• Only the described columns are provided. Participants must infer from provided text, images, or PDFs.
• Ensure submissions strictly adhere to the specified CSV formats.

Good luck!