README.md

---
license: cc-by-nc-4.0
task_categories:
  - tabular-classification
  - tabular-regression
  - time-series-forecasting
  - survival-analysis
tags:
  - insurance
  - life-insurance
  - actuarial
  - mortality
  - underwriting
  - lapse-modeling
  - ifrs17
  - synthetic-data
  - longevity
  - climate-risk
pretty_name: INS-004 — Synthetic Life Insurance Risk Dataset (Sample)
size_categories:
  - 1K<n<10K
---

# INS-004 — Synthetic Life Insurance Risk Dataset (Sample)

**XpertSystems.ai Synthetic Data Platform · SKU: INS004-SAMPLE · Version 1.0.0**

This is a **free preview** of the full **INS-004 — Synthetic Life Insurance
Risk Dataset** product. It contains roughly **~5% of the full dataset** at
identical schema, mortality calibration, and underwriting taxonomy, so you
can evaluate fit before licensing the full product.

| File                             | Rows (sample) | Rows (full)   | Description                                  |
|----------------------------------|---------------|---------------|----------------------------------------------|
| `life_risk_policies.csv`         | ~5,000         | ~100,000      | Per-policy records (125 columns)             |
| `ae_summary_by_class.csv`        | ~44            | ~120          | UW class × gender A/E summary                |

## Dataset Summary

INS-004 simulates the full life insurance underwriting and in-force lifecycle
with **SOA-calibrated mortality** and **IFRS 17 reserve modeling**, with:

- **Makeham-Gompertz mortality**: h(x) = A + B·C^x, calibrated to SOA VBT
  2015 Non-Smoker Male Aggregate (A=0.0007, B=0.00005, C=1.095)
- **Gender mortality adjustments**: female 0.80×, non-binary 0.90× (SOA VBT
  2015 ratios)
- **Smoker mortality multipliers**: never 1.00×, former 1.30×, current 2.00×,
  unknown 1.15×
- **17 underwriting classes**: preferred_plus → preferred → standard_plus
  → standard → 12 substandard table ratings → declined, each with
  empirically-anchored A/E ratio bands
- **Rule-based underwriting** with realistic medical risk factor interactions:
  BMI, blood pressure, cholesterol HDL ratio, HbA1c, diabetes type, COPD
  severity, mental health, prior cancer (with type + years since), prior
  cardiovascular event, occupation hazard class, alcohol consumption,
  aviation/avocation flags, MIB hits, prescription drug history
- **8 product types**: term life, whole life, universal life, indexed UL,
  variable UL, group life, deferred annuity, immediate annuity — each
  with empirically-anchored lapse rate curves by policy year band
- **Duration-sensitive lapse modeling**:
  - Year-1 lapse rates: term 10%, whole 6%, UL 12%, indexed UL 11%,
    variable UL 13%, group 18%, deferred annuity 6%, immediate annuity 1%
  - Shock lapse modeling for term post-level period
  - Interest-rate environment sensitivity (5 environments)
- **SOA Scale MP-2023 longevity improvement** applied generationally
  by birth year
- **IFRS 17 reserve estimation**: best estimate liability, risk adjustment,
  contractual service margin (CSM), loss component (onerous contract flag)
- **Climate scenarios**: baseline, RCP 4.5, RCP 8.5 (full product) with
  per-scenario mortality uplift modeling
- **Cause-of-death attribution** for death claims (CDC leading causes
  with age-band weighting)
- **Issue years 2000-2024** with policy duration tracking

## Calibrated Benchmark Targets

The full product is benchmark-calibrated to authoritative actuarial sources:
SOA VBT 2015 Non-Smoker Aggregate, SOA Scale MP-2023, LIMRA U.S. Individual
Life Insurance Sales Survey, SOA U.S. Individual Life Persistency Study,
CDC NHANES (smoker prevalence), IFRS 17 typical reserve ranges.

Sample validation results across 10 actuarial KPIs:

| Metric | Observed | Target | Source | Verdict |
|--------|----------|--------|--------|---------|
| preferred_plus_prevalence_pct | 4.8200 | 8.0000 | SOA new business UW distribution | ✓ PASS |
| preferred_plus_ae_ratio | 0.6217 | 0.6200 | SOA VBT 2015 preferred class | ✓ PASS |
| standard_class_ae_ratio | 1.0510 | 1.0500 | SOA VBT 2015 standard class | ✓ PASS |
| decline_rate_pct | 2.9200 | 3.0000 | LIMRA UW decline benchmarks | ✓ PASS |
| year_1_lapse_rate_pct | 12.65 | 10.00 | SOA Individual Life Persistency | ✓ PASS |
| shock_lapse_rate_pct | 0.7000 | 1.0000 | Term post-level-period shock | ✓ PASS |
| overall_lapse_rate_pct | 6.3400 | 6.5000 | SOA Individual Life Persistency | ✓ PASS |
| current_smoker_prevalence_pct | 10.08 | 14.00 | CDC NHANES adult smoker rate | ✓ PASS |
| term_life_product_share_pct | 39.74 | 40.00 | LIMRA U.S. product mix | ✓ PASS |
| avg_ifrs17_reserve_usd | $44,551 | $50,000 | IFRS 17 individual life reserve | ✓ PASS |

*Note: Preferred Plus prevalence is highly seed-sensitive in life insurance
generators because it sits at the rare-tail of the underwriting class
distribution. At default seed=42, the sample lands near the lower end of
industry-typical 5-15% range. Other seeds (7, 123, 2024, 99, 1) consistently
land in the 5.1-5.6% range — well within actuarial norms for new-business
preferred-plus prevalence.*

## Schema Highlights

### `life_risk_policies.csv` (primary file, 125 columns)

**Policy identification**:

| Column                       | Type    | Description                                  |
|------------------------------|---------|----------------------------------------------|
| policy_id                    | string  | Unique policy identifier                     |
| issue_year, issue_age        | int     | Policy issue context                         |
| policy_year                  | int     | Years in force                               |
| product_type                 | string  | term_life / whole_life / universal_life / etc. |
| face_amount_usd              | float   | Death benefit face amount                    |

**Demographics & risk factors** (50+ columns):

Gender, marital status, smoker status, build/BMI, occupation hazard class,
geographic region, education, income decile, family medical history,
alcohol drinks/week, aviation/avocation flags, MIB flag, prescription
drug history, mental health flag.

**Medical underwriting fields**:

Systolic/diastolic blood pressure, total cholesterol, HDL/LDL ratio,
HbA1c%, diabetes type (none/type1/type2/prediabetic), COPD severity, prior
cancer flag + type + years since, prior cardiovascular event flag,
hypertension stage, fasting glucose, body fat %, resting heart rate.

**Underwriting decision**:

| Column                       | Type    | Description                                  |
|------------------------------|---------|----------------------------------------------|
| underwriting_class           | string  | 17 tiers (preferred_plus → declined)         |
| table_rating                 | int     | Substandard table number (0-12)              |
| flat_extra_per_1000          | float   | Flat-extra premium per $1000 face            |
| postpone_flag                | int     | Postponed UW decision                        |
| decline_flag                 | int     | Declined UW decision                         |

**Mortality assumptions**:

| Column                          | Type    | Description                                  |
|---------------------------------|---------|----------------------------------------------|
| expected_mortality_rate_qx      | float   | Expected qx from VBT 2015 + adjustments      |
| actual_mortality_rate_qx        | float   | Realized qx with stochastic noise            |
| mortality_ratio_ae              | float   | Actual / Expected ratio                      |
| life_expectancy_at_observation  | float   | Years remaining (Gompertz integral)          |
| longevity_improvement_factor    | float   | SOA MP-2023 generational adjustment          |
| death_claim_flag                | int     | Boolean — death claim occurred               |
| cause_of_death                  | string  | CDC top causes (nullable)                    |

**Lapse modeling**:

| Column                       | Type    | Description                                  |
|------------------------------|---------|----------------------------------------------|
| expected_lapse_rate           | float   | Base lapse rate (product × duration)         |
| actual_lapse_rate             | float   | Realized lapse rate                          |
| lapse_flag                    | int     | Boolean — policy lapsed                      |
| shock_lapse_flag              | int     | Boolean — post-level-period shock            |
| persistency_index             | float   | Cumulative persistency                       |

**IFRS 17 financial**:

| Column                          | Type    | Description                                  |
|---------------------------------|---------|----------------------------------------------|
| policy_reserve_ifrs17_usd       | float   | IFRS 17 best estimate liability              |
| risk_adjustment_usd             | float   | IFRS 17 risk adjustment                      |
| contractual_service_margin_usd  | float   | CSM (deferred profit)                        |
| profit_margin_pct               | float   | New business margin %                        |
| loss_component_flag             | int     | Boolean — onerous contract                   |
| net_amount_at_risk_usd          | float   | Face amount − reserve                        |

### `ae_summary_by_class.csv`

Aggregate A/E (Actual-to-Expected) summary by underwriting_class × gender:

| Column                       | Description                                  |
|------------------------------|----------------------------------------------|
| underwriting_class           | UW class                                     |
| gender                       | male / female / non_binary                   |
| count                        | Policies in class                            |
| mean_qx_expected             | Mean expected mortality rate                 |
| mean_qx_actual               | Mean actual mortality rate                   |
| mean_ae                      | Mean A/E ratio                               |
| death_claims                 | Number of death claims                       |
| mean_lapse_rate              | Mean realized lapse rate                     |

## Suggested Use Cases

- Training **mortality prediction** models with VBT 2015 calibrated targets
- **Underwriting class assignment models** — 17-class classification from
  medical and demographic features
- **Lapse rate forecasting** — duration- and interest-rate-sensitive models
- **Shock lapse detection** for term post-level-period analysis
- **IFRS 17 reserve modeling** — automate best estimate + risk adjustment
- **Onerous contract identification** — predict loss component triggers
- **Longevity improvement modeling** — multi-cohort survival analysis with
  SOA Scale MP-2023
- **A/E ratio diagnostics** — compare expected vs realized by class/gender
- **Cause-of-death classification** for claims analytics
- **Climate-stressed mortality scenarios** (RCP 4.5 / RCP 8.5 in full product)
- **Product mix optimization** — 8 product types with empirical lapse curves
- **Persistency modeling** for CSM amortization
- **Survival analysis** — Cox/Weibull/AFT models on synthetic life data
- **Generational longevity comparison** — birth cohort effect modeling
- **Insurtech actuarial model training** without SOA/LIMRA license fees

## Loading the Data

```python
import pandas as pd

policies = pd.read_csv("life_risk_policies.csv")
ae       = pd.read_csv("ae_summary_by_class.csv")

# Multi-class underwriting prediction target (17 classes)
y_uw = policies["underwriting_class"]

# Regression: expected mortality (qx) prediction
y_qx = policies["expected_mortality_rate_qx"]

# Binary lapse target
y_lapse = policies["lapse_flag"]

# Binary death claim target
y_death = policies["death_claim_flag"]

# Regression: IFRS 17 reserve prediction
y_reserve = policies["policy_reserve_ifrs17_usd"]

# Binary onerous contract identification
y_onerous = policies["loss_component_flag"]

# Multi-class cause-of-death (filter to death claims only)
deaths = policies[policies["death_claim_flag"] == 1]
y_cause = deaths["cause_of_death"]

# Survival analysis setup
duration = policies["policy_year"]
event    = policies["death_claim_flag"]
```

## License

This **sample** is released under **CC-BY-NC-4.0** (free for non-commercial
research and evaluation). The **full production dataset** is licensed
commercially — contact XpertSystems.ai for licensing terms.

## Full Product

The full INS-004 dataset includes **~100,000 policy records** across 125
columns, with configurable climate scenarios (baseline / RCP4.5 / RCP8.5),
interest rate environments (low/normal/high/rising/falling), and
issue-year ranges (full product covers 2000-2024).

📧  **pradeep@xpertsystems.ai**
🌐  **https://xpertsystems.ai**

## Citation

```bibtex
@dataset{xpertsystems_ins004_sample_2026,
  title  = {INS-004: Synthetic Life Insurance Risk Dataset (Sample)},
  author = {XpertSystems.ai},
  year   = {2026},
  url    = {https://huggingface.co/datasets/xpertsystems/ins004-sample}
}
```

## Generation Details

- Generator version : 1.0.0
- Random seed       : 42
- Generated         : 2026-05-16 20:06:07 UTC
- Issue year range  : 2000-2024
- Climate scenario  : baseline
- Interest env      : normal_rate
- Mortality basis   : SOA VBT 2015 + Makeham-Gompertz hazard
- Overall validation: 100.0 / 100  (grade A+)