Auto Insurance Pricing Model

This project develops a sophisticated pricing model for auto insurance using statistical and machine learning techniques. It includes exploratory data analysis, feature engineering, frequency and severity modeling, and risk-adjusted premium calculation.

Key Features

• Exploratory Data Analysis (EDA) of claim frequency and severity
• Advanced feature engineering to create novel risk factors
• Frequency modeling using Poisson, Negative Binomial, and Zero-Inflated Poisson models
• Severity modeling using log-normal regression
• Risk-adjusted premium calculation system
• Visualization of pricing impacts and risk factor analysis

Technologies Used

• R
• Libraries: data.table, dplyr, ggplot2, MASS, pscl, caret

Project Structure

1. EDA.R: Exploratory Data Analysis
2. Feature_engineering.R: Creation of new risk factors
3. Frequency_model.R: Claim frequency modeling
4. Severity_model.R: Claim severity modeling
5. Risk_analysis.R: Risk factor analysis and premium calculation

Risk Analysis Visualizations

Feature Engineering Impact Analysis

This visualization demonstrates the relationship between different risk factors and claim frequency. Key insights:

• Cars aged 6-10 years show the highest claim frequency
• Young drivers (18-25) have significantly higher claim rates
• Higher population density correlates with increased claim frequency
• Full year policies show different risk patterns compared to partial year coverage

Risk Heat Map Analysis

The heat map reveals the interaction between driver age and vehicle age groups:

• Highest risk concentration (red) appears in young drivers (18-25) with vehicles aged 6-10 years
• More experienced drivers (45+) with newer vehicles show lower claim rates (blue)
• Clear pattern of risk reduction as driver age increases
• Vehicle age has a non-linear impact on risk across different driver age groups

Population Density Risk Analysis

Analysis of risk patterns across population density groups shows:

• Medium-High density areas have the highest risk score (2.45)
• Clear correlation between population density and risk
• Risk scores range from 1.80 to 2.45, showing significant variation
• The relationship is not perfectly linear, suggesting other factors influence risk in urban vs. rural areas

Results

• 15% improvement in premium accuracy compared to baseline model
• Increase in model predictive power through novel risk factors

Future Work

• Implement Generalized Additive Models (GAMs) for non-linear relationships
• Explore machine learning approaches like Random Forests or Gradient Boosting Machines
• Conduct competitive analysis and market basket analysis