Medical Insurance Charges Bayesian Analysis Dataset

Overview

The purpose of this dataset is to develop optimal health‑insurance products by leveraging historical records to estimate individual medical costs. These data support the creation of precise pricing models, strategic insurance planning, and effective portfolio management. The main goal is to accurately predict insurance charges using a variety of predictor variables.

Dataset Details

• Number of Records: 1,339 health‑insurance entries
• Target Variable: Individual medical charges ("Charges")
• Features:
  • Age (numeric): Age of the policyholder
  • Sex (categorical): Gender of the policyholder (male or female)
  • BMI (numeric): Body Mass Index (kg/m²)
  • Children (numeric): Number of children covered
  • Smoker (categorical): Whether the policyholder smokes
  • Region (categorical): Beneficiary’s U.S. region (Northeast / Southeast / Southwest / Northwest)
  • Charges (numeric): Individual medical cost billed by the health‑insurance company

Note: Synthetic data were used to preserve privacy while maintaining a 95% similarity to the original dataset, ensuring compliance with GDPR and related regulations.

Required Libraries

The following R libraries are used for analysis:

c("dplyr", "MCMCpack", "coda", "R2OpenBUGS", "mixAK", "brms")

Pre‑processing

The original dataset underwent several modifications, such as converting categorical variables to binary indicators and reformatting certain fields to facilitate analysis.

Models Employed

Linear Models

• Markov Chain Monte Carlo (MCMC) sampling
• OpenBUGS
• Frequentist Generalized Linear Model (GLM)
• Bayesian GLM
• Normal‑mixture MCMC

Non‑linear Models

• MCMC sampling
• Bayesian Generalized Additive Model (GAM) (non‑linear)

Conclusion

Detailed conclusions and results are presented in the accompanying file, including figures and additional relevant information.