This repo reproduces the paper Classification with Strategically Withheld Data. The paper appeared in AAAI 2021, GAIW 2021 and IML 2020. This repo requires Python 3.7.

3 of the 4 datasets are directly retrieved from the internet. One of the dataset needs to be downloaded from the UCI ML Repo.

To run the experiment

python credit.py <dataset> <fraction> <balanced-flag>

where <dataset> can be "australia", "germany", "poland" or "taiwan", specifying the dataset to run the experiment on; <fraction> can be any value between 0 and 1 (in our work, we do 0.0, 0.1, 0.2, 0.3, 0.4, 0.5), specifying the fraction of data missing for non-strategic reasons; <balanced-flag> can be "bal" or blank, specifying whether or not to balance the dataset before experiement.

To reproduce the paper, run with each of the 4x6x2=48 specifications.

To generate all tables and figures in tex and png:

python parse.py