To help combat discrimination in mortgage lending, researchers at the Massachusetts Institute of Technology developed a process to remove bias from the data used to train machine learning (ML) models.
The technique, called DualFair, subdivides a dataset into the largest number of subgroups based on combinations of sensitive attributes and options to eliminate label bias.
DualFair evens out the number of borrowers in each subgroup by duplicating individuals from minority groups and deleting individuals from the majority group, then balances the proportion of loan acceptances and rejections in each subgroup to match the median in the original dataset before recombining them.
To eliminate selection bias, DualFair iterates on each datapoint to identify discrimination, removing those found to be biased from the dataset.
The researchers found their method lowered discrimination in predictions, while maintaining high accuracy.
From MIT News
View Full Article
Abstracts Copyright © 2022 SmithBucklin, Washington, DC, USA