RigbeWeleslasie/insurance-risk-analytics

Predictive risk modeling and pricing optimization using historical South African auto-insurance data. Implements EDA, A/B test validation, DVC, and SHAP interpretability. # Insurance Risk Analytics – ACIS Predictive risk modelling and pricing optimisation for AlphaCare Insurance Solutions (ACIS), using 18 months of historical South African auto-insurance claim data (Feb 2014 – Aug 2015). ## Business Context ACIS is preparing for aggressive growth in the SA auto-insurance market. This project provides the analytical foundation for: - Identifying **low-risk client segments** where premiums can be reduced to attract new business. - Statistically validating **risk hypotheses** across provinces, zip codes, and demographics. - Building **predictive models** for claim probability and severity that power dynamic, risk-based pricing. ## Project Structure ``` insurance-risk-analytics/ ├── .github/workflows/ci.yml # GitHub Actions CI (lint + test on every push) ├── data/ # Tracked by DVC — not committed to Git ├── notebooks/ │ ├── 01_eda.ipynb # Exploratory Data Analysis (Task 1) │ ├── 02_hypothesis_testing.ipynb # A/B hypothesis tests (Task 3) │ └── 03_modeling.ipynb # Predictive modelling (Task 4) ├── src/ │ ├── data_loader.py # DataLoader class — load, validate, enrich │ ├── eda_utils.py # EDA helper functions and plots │ ├── hypothesis_tests.py # Statistical test utilities (Task 3) │ └── modeling.py # Model training and evaluation (Task 4) ├── reports/ │ └── final_report.md # Medium-style business report ├── tests/ # pytest test suite ├── .dvc/ # DVC config (committed) ├── dvc.yaml # DVC pipeline stages ├── requirements.txt └── README.md ``` ## Key Metrics | Metric | Formula | Purpose | |--------|---------|---------| | **Loss Ratio** | `TotalClaims / TotalPremium` | Core measure of portfolio profitability | | **Margin** | `TotalPremium - TotalClaims` | Per-policy profit contribution | | **Claim Frequency** | `count(claims > 0) / count(policies)` | Probability of a claim occurring | …