Fairness Aware Counterfactuals for Subgroups
Authors: Loukas Kavouras, Konstantinos Tsopelas, Giorgos Giannopoulos, Dimitris Sacharidis, Eleni Psaroudaki, Nikolaos Theologitis, Dimitrios Rontogiannis, Dimitris Fotakis, Ioannis Emiris
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the advantages, the wide applicability, and the efficiency of our approach through a thorough experimental evaluation of different benchmark datasets. |
| Researcher Affiliation | Academia | Loukas Kavouras IMSI / Athena RC kavouras@athenarc.gr Konstantinos Tsopelas IMSI / Athena RC k.tsopelas@athenarc.gr Giorgos Giannopoulos IMSI / Athena RC giann@athenarc.gr Dimitris Sacharidis ULB dimitris.sacharidis@ulb.be Eleni Psaroudaki NTUA & IMSI / Athena RC epsaroudaki@mail.ntua.gr Nikolaos Theologitis IMSI / Athena RC n.theologitis@athenarc.gr Dimitrios Rontogiannis IMSI / Athena RC dronto@gmail.com Dimitris Fotakis NTUA & Archimedes / Athena RC fotakis@cs.ntua.gr Ioannis Emiris NKUA & IMSI / Athena RC emiris@athenarc.gr |
| Pseudocode | No | The paper describes the steps of the FACTS framework in narrative text but does not include structured pseudocode or a clearly labeled algorithm block. |
| Open Source Code | Yes | The code is available at: https://github.com/AutoFairAthenaRC/FACTS. |
| Open Datasets | Yes | This section presents the experimental evaluation of FACTS on the Adult dataset [1]; more information about the datasets, experimental setting, and additional results can be found in the appendix. The code is available at: https://github.com/AutoFairAthenaRC/FACTS. ... C Datasets Description ... 7https://raw.githubusercontent.com/columbia/fairtest/master/data/adult/adult.csv 8https://aif360.readthedocs.io/en/latest/modules/generated/aif360.sklearn. datasets.fetch_compas.html 10https://raw.githubusercontent.com/samuel-yeom/fliptest/master/exact-ot/ chicago-ssl-clean.csv 11https://developer.ibm.com/exchanges/data/all/bias-in-advertising/ 12https://archive.ics.uci.edu/dataset/144/statlog+german+credit+data |
| Dataset Splits | No | The resulting dataset was split randomly with a 70:30 split ratio and was used to train and test respectively a logistic regression model (consequently used as the black-box model to audit). |
| Hardware Specification | Yes | Experiments were run on commodity hardware (AMD Ryzen 5 5600H processor, 8GB RAM). |
| Software Dependencies | No | To conduct our experiments, we have used the Logistic Regression classification model, where we use the default implementation of the python package scikit-learn. ... Our results can be reproduced using the random seed value 131313 in the data split function (train_test_split from the python package scikit-learn). ... We used the implementation in the Python package mlxtend. ... On the software side, all experiments were run in an isolated conda environment using Python 3.9.16. |
| Experiment Setup | Yes | Experimental Setting The first step was the dataset cleanup (e.g., removing missing values and duplicate features, creating bins for continuous features like age). The resulting dataset was split randomly with a 70:30 split ratio and was used to train and test respectively a logistic regression model... For the generation of the subgroups and the set of actions we used fp-growth with a 1% support threshold on the test set. We also implemented various cost functions, depending on the type of feature, i.e., categorical, ordinal, and numerical. ... We used two different values arbitrarily, i.e., a relatively low effectiveness level of ϕ = 30% and a relatively high effectiveness level of ϕ = 70%. For the estimation of budget-level values c we followed a more elaborate procedure. ... We have chosen the 30%, 60% and 90% percentiles arbitrarily. |