Insurance Customer Risk Segmentation & Pricing Optimization

Goal: Use synthetic insurance data to simulate claims, discover risk profiles, and optimize pricing via unsupervised machine learning.
Tech Stack: Python, Pandas, Scikit-learn, Seaborn, PCA + KMeans
Key Outcome: 4 actionable customer clusters with pricing recommendations
License: MIT

---

Project Overview

This project demonstrates a full data science pipeline for insurance risk analytics:

1. Raw synthetic customer data → data/
2. Monte Carlo claim simulation + clustering → src/
3. Visualizations + actionable insights → results/

The result: 4 customer risk segments that reveal underpriced policies, high-risk losers, and profit drivers.

---

Project Structure

insurance-risk-segmentation/
│
├── data/                  # Raw input data
│   ├── data_synthetic.csv
│   └── README.md          # Dataset description
│
├── src/                   # Analysis code
│   ├── analysis_pipeline.py
│   └── README.md          # Step-by-step pipeline explanation
│
├── results/               # Outputs & insights
│   ├── cluster_summary.csv
│   ├── clustering_pca_2d.png
│   ├── frequency_histogram.png
│   ├── expected_loss_histogram.png
│   ├── severity_histogram.png
│   ├── freq_vs_sev_scatter.png
│   └── README.md          # Full results & business actions
│
└── README.md              # This file

---

Key Results

Cluster	Name	Risk	Avg. Premium	Expected Loss	Profit	Action
0	Gold Safe	Low	3,223 €	19.71 €	+3,203 €	Retain & upsell
1	High Frequency Loser	Very High	3,539 €	425 €	+3,114 €	Increase premium or exclude
2	Balanced Premium	Medium-High	3,790 €	147 €	+3,643 €	Maintain & reward
3	Claim Heavy, Low Premium	High	1,939 €	47.91 €	+1,892 €	Increase to 3.2k+

Critical Insight:
Cluster 3 has Claim History = 3.71 (near max) but pays only 1.9k → underpriced by ~60%.

---

Visual Highlights

Plot	Insight
	Costumer clustering: 4 clear, compact groups → interpretable risk segments
	Frequency distribution: Right-skewed most have 0–1 claims
	Loss distribution: Long tail - few high-cost customers dominate risk
	Frequency vs Severity: Strong correlation; Risk Profile drives upper tail

See all plots in results/

---

How to Run

# 1. Clone the repo
git clone https://github.com/yourname/insurance-risk-segmentation.git
cd insurance-risk-segmentation

# 2. Install dependencies
pip install pandas numpy matplotlib seaborn scikit-learn

# 3. Run the analysis
python src/analysis_pipeline.py

Outputs saved automatically to results/

---

Business Impact

Action	Estimated Benefit
Reprice Cluster 3 (+1.3k/customer)	+65 M€ (50k customers)
Exclude or penalize Cluster 1	Reduce loss exposure
Upsell Cluster 0 & 2	Increase retention & revenue

---

Reproducibility

• Deterministic: random_state=42 in PCA, KMeans, and simulation
• Pipeline-based: Consistent preprocessing
• Documented: Full READMEs in every folder

---

Author

[Salvatore Spagnuolo]