mubarek4566/AlphaCare_car_insurance
Domain:
socioeconomicRecord type:
project
Creator:
mub
Host:
This repository contains analyses and models to support AlphaCare Insurance Solutions' marketing strategy using historical car insurance claim data from South Africa (Feb 2014 - Aug 2015).
# AlphaCare Car Insurance Risk Analytics
This repository showcases predictive models for optimizing car insurance marketing and risk management at AlphaCare Insurance Solutions (ACIS). Using historical claim data, it identifies trends, low-risk clients, and opportunities to reduce premiums, enhancing customer acquisition and retention through data-driven strategies.
## Objectives
- Identify low-risk customers for premium optimization
- Conduct A/B hypothesis testing
- Build predictive models for premium and claim prediction
- Provide actionable insights for marketing decisions
## Tech Stack
- Python
- Pandas, Scikit-learn, Statsmodels, Seaborn, Matplotlib
- Git, GitHub
- GitHub Actions (CI/CD)
## 🗃️ Data Version Control (DVC)
To ensure reproducibility and compliance in a regulated industry like insurance, DVC is used to version-control datasets and pipeline artifacts.
### Steps to Reproduce:
1. Install DVC:
```bash
pip install dvc
Initialize DVC:
dvc init
Add Local Remote Storage:
mkdir -p ./dvc-storage
dvc remote add -d localstorage ./dvc-storage
Track and Version Raw Dataset:
dvc add data/raw/insurance_data.csv
git add data/raw/insurance_data.csv.dvc .gitignore
git commit -m "Track raw dataset with DVC"
Push Dataset to Local Remote:
dvc push
# ✅ Model implementation:
## Data Preparation:
Handles missing data with median imputation for numerical features and 'Unknown' for categorical features
Creates new features like vehicle age, coverage ratio, and power-to-capacity ratio
Converts dates to datetime and extracts temporal features
## Model Building:
💡 Claim severity model (Random Forest Regressor) trained only on policies with claims
💡 Claim probability model (Random Forest Classifier) trained on all policies
💡 Uses pipelines with preprocessing for both numerical and categorical features
## Evaluation:
💡 For regression: RMSE and R-squared
💡 For classification: AUC and accuracy
## Risk-Based Pricing:
💡 Implements the formula: Premium = (Predic …