nathanaeldereje/ACIS-insurance-risk-analytics

End-to-end car insurance risk analytics for AlphaCare Insurance (South Africa). EDA, A/B hypothesis testing (province, postal code, gender), loss-ratio analysis, DVC pipelines, and XGBoost + SHAP models to identify low-risk segments and optimize premiums. # ACIS Insurance Risk Analytics & Predictive Pricing (South Africa) **AlphaCare Insurance Solutions – Motor Insurance Analytics Challenge | Dec 2025** End-to-end risk analytics project to identify low-risk customer segments, statistically validate risk drivers, and build a predictive pricing engine that enables targeted premium reductions for the South African car insurance market. ### Business Goal Help AlphaCare Insurance Solutions (ACIS) attract profitable new customers by: - Discovering low-risk provinces, postal codes, driver profiles, and vehicle types - Proving where statistically significant risk differences exist (and where they don’t) - Building interpretable ML models for claim prediction and optimal risk-based premiums ### Key Deliverables - Comprehensive EDA + beautiful, insight-driven visualizations - Rigorous A/B hypothesis testing (provinces, postal codes, gender, margins) - Reproducible data pipeline with **DVC** (Data Version Control) - Claim probability + severity models → risk-based premium framework - XGBoost + SHAP explanations for top risk drivers - Concrete marketing & pricing recommendations backed by p-values and feature impacts ### Project Structure ```bash ACIS-insurance-risk-analytics/ ├── data/ # Raw + versioned datasets (tracked with DVC) │ ├── raw/ │ │ └── MachineLearningRating_v3.txt.dvc # tracked by DVC │ └── processed/ # future cleaned versions ├── notebooks/ # Exploratory analysis & final visualizations ├── scripts/ # Clean, modular production pipeline ├── reports/ # Interim + final report (PDF/Medium-style) ├── sql/ # Optional queries & aggregations ├── .dvc/ # DVC configuration & cache ├── .gitignore ├── requirements.txt └── README.md ``` ### Tech Stack Python • Pandas • Scikit-learn • XGBoost • SHAP • Matplotlib/Seaborn • Plotly • DVC • GitHub Actions • Jupyter ### Quick Start ```bash git cl …