JultikaUniversity of Oulu repository

Abstract

As the usage and impact of machine learning applications increase, it is increasingly important to ensure that the systems in use are beneficial to users and larger society around them. One of steps to ensure this is limiting unfairness that the algorithm might have. Existing machine learning applications have sometimes shown that they have been disadvantageous to certain minorities and to combat this we have a need for defining what does fairness means, and how can we increase it in our machine learning applications. The survey is done as a literary review with the goal of presenting an overview of fairness in classification-based machine learning. The survey goes through the motivation for fairness briefly through philosophical background and examples of unfairness and goes through the most popular fairness definitions in machine learning. After this the paper lists some of the most important methods for restricting unfairness splitting the methods into pre- in- and post-processing methods.