fairness-aware-classification

Skoltech ML Course project "Boosting for Fairness-Aware Classification"

Compared methods

Method	Description
SMOTE + LogisticRegression	SMOTE: Synthetic Minority Over-sampling Technique
SMOTE + RandomForest	SMOTE: Synthetic Minority Over-sampling Technique
SMOTEBoost	SMOTEBoost: Improving Prediction of the Minority Class in Boosting
RUSBoost	RUSBoost: A Hybrid Approach to Alleviating Class Imbalance
AdaBoost	A Decision-Theoretic Generalization of On-Line Learning and an Application to Boosting
AdaFair	AdaFair: Cumulative Fairness Adaptive Boosting
AttentiveGradientBoosting	Our novel approach

How to run

1. Firstly to install all required packages from requirements.txt you should run pip install -r requirements.txt.
2. Secondly you need to download all datasets. To do it you should just run script download.sh from working directory. Note, that kaggle pip library should be installed to download some datasets. As a result four datasets will be downloaded and folders adult, bank, compass, kdd will be created.
3. To run experiments you can just run notebook experiments.ipynb. As a result plots will be saved in folder plots.

Description of the structure

• requirements.txt: list of packages that should be installed

• download.sh: script for downloading datasets

• download.ipynb: notebook for downloading datasets (similar to script)

• experiments.ipynb: notebook to run experiments

• algo.py: implementation of all algorithms are written here

• utils.py: functions for loading and preprocessing datasets are written here

• metrics.py: implementations of all required metrics are written here

• presentation.pdf: presentation of the project

• report.pdf: report of the project

• plots: folder with plots created after the experiment