Skip to content

This is a Technical Audit on Prudential's ADS system as part of a Kaggle Competition by Amando Xu and Rohan Mukerji

Notifications You must be signed in to change notification settings

PrettyAX/Prudential-ADS-Technical-Audit

Repository files navigation

Prudential Automated Decision System (ADS) Technical Audit

Overview

This project presents a comprehensive technical audit of Prudential's Automated Decision System (ADS), which uses advanced data analytics and machine learning to revolutionize life insurance risk assessment. Developed in response to a Kaggle competition by Prudential, the ADS implementation leverages ridge regression to predict risk scores for life insurance policies.

Objective

The ADS aims to streamline the life insurance application process, which traditionally involves extensive client information gathering, medical exams, and long waiting periods for quotes. Our solution offers a more efficient, less labor-intensive, and unbiased method for obtaining life insurance quotes, aligning with modern expectations of instant service and data privacy standards.

Implementation

  • Input Data: Utilizes a dataset of 59,381 individual records with 128 columns covering demographic, employment, insurance history, family history, and medical information.
  • Modeling Approach: Employs Ridge Regression models for prediction, focusing on minimizing the mean squared error (MSE) and ensuring fair and unbiased assessments across various sub-populations.
  • Performance Evaluation: Analyses include Root Mean Square Error (RMSE) metrics across different subpopulations (age, height, weight, BMI) and fairness assessments using demographic parity and equalized odds metrics.

Outcomes

  • The model exhibits varying performance across different subpopulations, indicating the need for further research and improvements.
  • Demographic parity analysis reveals disparities in positive prediction rates across subpopulations, highlighting fairness issues in the ADS.

Future Directions

  • Employ advanced methods like SHAP (SHapley Additive exPlanations) to better understand model predictions, identify potential biases, and improve performance.
  • Explore more complex model selections to enhance robustness and accuracy.

Authors

Acknowledgments

Special thanks to Prudential and the Kaggle community for providing the data and platform for this innovative project.

About

This is a Technical Audit on Prudential's ADS system as part of a Kaggle Competition by Amando Xu and Rohan Mukerji

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published