FenetShewarega/insurance-risk-analytics
Domaine:
socioeconomicType de record:
project
Créateur:
Fen
Hôte:
End-to-end insurance risk analytics & predictive modeling — ACIS South African auto-insurance (AlphaCare Insurance Solutions)
# Insurance Risk Analytics & Predictive Modeling
End-to-end analytics project for **AlphaCare Insurance Solutions (ACIS)**: analyse 18 months of South African auto-insurance claim data (Feb 2014 – Aug 2015), validate risk hypotheses, and build risk-based pricing models.
## Business Context
ACIS is preparing for an aggressive growth phase in the South African auto-insurance market. The goals are to:
1. Identify **low-risk customer segments** where premiums can be reduced to attract new clients.
2. **Statistically validate** hypotheses about risk drivers (province, zip code, gender).
3. Build **predictive models** for claim severity and claim probability that feed a dynamic, risk-based premium.
4. Deliver clear **business-facing recommendations**.
## Key Metrics
- **Loss Ratio** = `TotalClaims / TotalPremium` — portfolio profitability.
- **Margin** = `TotalPremium − TotalClaims` — per-policy profit contribution.
- **Claim Frequency** — proportion of policies with at least one claim.
- **Claim Severity** — mean claim amount given a claim occurred.
## Project Structure
```
insurance-risk-analytics/
├── .github/workflows/ci.yml # Lint + tests on every push
├── data/ # DVC-tracked, not in Git
├── notebooks/
│ ├── 01_eda.ipynb
│ ├── 02_hypothesis_testing.ipynb
│ └── 03_modeling.ipynb
├── src/ # Reusable Python modules
│ ├── data_loader.py
│ ├── eda_utils.py
│ ├── hypothesis_tests.py
│ └── modeling.py
├── reports/final_report.md
├── tests/
├── dvc.yaml
├── requirements.txt
└── README.md
```
## Setup
```bash
# 1. Clone & enter the repo
git clone && cd insurance-risk-analytics
# 2. Create a virtual environment
python -m venv .venv && source .venv/bin/activate
# 3. Install dependencies
pip install -r requirements.txt
# 4. Pull the data from the DVC remote
dvc pull
```
## Reproducing the Data Pipeline (DVC)
Data is versioned with **DVC** so every analysis is reproducible.
```bash
# 1) Install …