Fairness Aware Classification Algorithms Report

About The Project • Built With • Download • Usage • License • Other Projects • Contact

Table of Contents

About The Project
Usage
License
Other Projects
Contact

About The Project

Introduction

Fairness-aware algorithms are designed to analyze data without being too bias on the result. The model algorithm is altered to lose dependence on A_s so that the data-set will be processed differently without losing accuracy in the number of attributes. Normally the data-set will process $S$ from A_s through the data-set and produce the result S, to lose the results dependence on A_s the model splits into two (M_+ and M_-). M_+ takes the favour values of S_+ whereas M_- takes the discriminated values S_- to which the final classifier will be dependent on S, where All the values in the tree (array) are associated with S, this is applicable for both M_+ and M_- as both share an identical Naïve Bayes model, the 2 Naïve models related to the M_+ and M_- models.

In this model specifically as stated takes all the discriminated values and assigns them to the M_- model based off of the S_- values. The algorithm is a base Naïve model algorithm which depends on a class C, instead of this class its modified where the S values are used. The S_- discriminated values are used that removes this discrimination from the Naïve Bayes algorithm.

Though both the fairness-aware algorithms are similar in the output goal, in which they process data into non discriminatory. However they are vastly different with the decision tree algorithm allowing for further usage and on larger more complex data-sets than that of the proposed 2 Naïve Bayes algorithm. Such examples would include that the Naïve Bayes model is useful for gender discrimination whereas the decision tree would look at age as it will be able to create new branches base off of the splitting criterion. Gender seeks only male or female, the data is split into two models as M_+ and M_-, where age seeks not only up to 100 values but also groups of age brackets, to which can be branched off the tree where the entropy of the splitting criterion will justify the next node.

Built With

LaTeX, Overleaf, BibTeX

Other Infomation

LaTeX Report: https://github.com/Mysftz/fairness-aware-classification-algorithms-report

(back to top)

Usage

This report was submitted for the degree of a Masters of Science in Computer Science (Artificial Intelligence) at the University of Kent in March 2022.

(back to top)