Limiting Bias Drift using Generative Adversarial Network Framework

This project will outline step-by-step how to make fair financial models that do not depend on a person’s private information such as age, gender and race. It is aimed to be a guide to use machine-learning tools and adding defendable mathematical theory to improve previously existing models that have problems with producing biased results. As such, our final goal of this project is for individuals to be fairly evaluated based on relevant and unbiased decision-making processes when applying to receive any form of financial support. Additionally, this framework will ensure reduction of computational efficiency while maximizing accuracy. We wish to have our framework to be the prototypical foundation to future development of existing decision-making financial models.

Faculty Supervisor:

Michael Kouritzin

Student:

Partner:

Scotiabank

Discipline:

Mathematics

Sector:

Finance and Insurance

University:

University of Alberta

Program: