dawit-hopes/insurance-risk-segmentation-ml

End-to-end insurance risk analytics project using real claims data from South Africa. Includes A/B testing, risk segmentation, statistical modeling, and machine-learning models to identify low-risk customer segments and optimize premium pricing. ## 🚗 End-to-End Insurance Risk Analytics & Predictive Modeling This repository, is a comprehensive data science project aimed at solving a critical business problem for **AlphaCare Insurance Solutions (ACIS)** in South Africa: **optimizing car insurance pricing and marketing by identifying low-risk client segments.** --- ## 💡 What This Project Does The project functions as an **End-to-End Risk Analytics and Predictive Modeling pipeline** using historical car insurance claim data (Feb 2014 – Aug 2015). Its core purpose is to transform raw policy data into **actionable business strategies** through the following sequential steps: ### 1. Risk and Profitability Analysis (EDA & Statistics) * **Identifies key drivers of risk** by analyzing the **Loss Ratio** and performing **A/B Hypothesis Testing** on major factors like **Province**, **Zip Code**, and **Gender**. * **Goal:** Statistically validates which demographic, geographic, or vehicle features lead to significant differences in **Claim Frequency**, **Claim Severity**, and **Profit Margin**. ### 2. Auditable Data Pipeline (DVC) * **Ensures reproducibility** for auditing and regulatory compliance (essential in the financial sector) by implementing **Data Version Control (DVC)**. * **Goal:** Rigorously tracks and versions the large historical dataset alongside the code, allowing any analysis or model result to be recreated precisely. ### 3. Predictive Pricing Models (Machine Learning) * **Develops advanced machine learning models** (including **XGBoost** and **Random Forests**) to forecast the financial liability associated with an insured policy. * **Modeling Focus:** Building a **Claim Severity Model** (predicting `TotalClaims`) that can be integrated into a formula to recommend **optimal, risk-based premium values**. ### 4. Model Interpretability * **Provides business context** for the predictive models using techniques like **SHAP** (SHapley Additive exPlanations). * **Goal:** Explains *why* the model make …