Bias in Machine Learning Models and Corrective Methods

All models but in particular Machine learning (ML) models run the risk of incorporating bias or unfairness in their outcomes. This is often driven by the underlying data used to train or calibrate the model. Since these models are not only increasingly used to make important decisions in our financial lives but also in other aspects like granting university admission, social benefit assignment, predicting risk of criminal recidivism, and part of hiring tools to review job applicant’s resumes, these biases have social, ethical as well as legal implications. This lecture gives a brief overview of the definitions of fairness used in the industry and some of the methods used to correct for unfairness in ML models.