Authors: Guilherme Dean Pelegrina, Renan Del Buono Brotto, Leonardo Tomazeli Duarte, Romis Attux, João Marcos Travassos Romano.
This work verifies the compromise between the (total) reconstruction error and the fairnes measure in a dimensional reduction problem. Fairness measure is given by the difference between the reconstruction errors of the two considered classes. We use a multi-objective approach (SPEA2 algorithm) and select a single non-dominated solution based on the minimum weighted sum (with equal importance, but taking the scales of each objective into account). We consider the Default Credit dataset - see [Yeh, I. C., & Lien, C. H. (2009). Expert Systems with Applications, 36(2), p. 2473-2480] - and the Labeled Faces in the Wild (LFW) - see [Huang, G. B., Mattar, M., Berg, T., & Learned-Miller, E. (2008). Labeled faces in the wild: A database for studying face recognition in unconstrained environments. In Workshop on Faces in 'Real-Life' Images: Detection, Alignment, % and Recognition. Marseille, France].
The comparisons of data mining techniques for the predictive accuracy of probability of default of credit card clients and some functions were based on [Samadi et al. (2018). The price of fair pca: One extra dimension. In Advances in Neural Information Processing Systems, p. 10976-10987].
To cite this work: Pelegrina, G. D.; Brotto, R. D. B.; Duarte, L. T.; Attux, R. & Romano, J. M. T. (2021). A novel multi-objective-based approach to analyze trade-offs in Fair Principal Component Analysis. ArXiv preprint, arXiv:2006.06137. Available at: https://arxiv.org/abs/2006.06137
All the files in this repository are in .m format, so it is necessary to execute them in a Matlab (v. 2015a) or Octave Environment
- Clone the repository
- Execute the file "main_moofpca_spea2_allMeth_credit.m" (modify the Load Data, if necessary)
All data files can be downloaded at https://drive.google.com/drive/folders/1ltUvPAj5rrBZO_pOl4N0QHXD_JdFmsE7?usp=sharing