Predictive model that estimates the severity of insurance claims of XXX

Goals

1. Cross-validate regression/emsemble models
2. Feature engineering, encoding, handling missing values
3. Document trade-offs between model complexity vs. interpretability
4. Summary deck w/ findings
5. Analyze feature importance and recommend what data matters most
6. Handling diverse feature scope (multi-columns)

Scenarios

• False Negative (missed claim): Risky policyholder charged standard rate → Company loses money on claims
• False Positive (wrongly flagged): Safe policyholder charged higher premium → Might lose customer to competitor

Let's assume:

• Average claim cost: $5,000
• Premium increase for flagged customers: $500/year
• Customer churn rate if wrongly flagged: 20%
• Lifetime value of lost customer: $2,000