The projects consists in taking the base model showed in "Obermeyer, Z., Powers, B., Vogeli, C., Mullainathan, S., Dissecting racial bias in an algorithm used to manage the health of populations, Science 366, https://escholarship.org/content/qt6h92v832/qt6h92v832.pdf, pp.447-453, 2019" and applying some methods for debiasing known in the literature. A comparison is then carried out between the models, with a correlated discussion on tradeoffs and accuracy. The models developed are:
- Decorrelated model (FairLearn Decorrelator)
- Augmented model (ADASYN)
- Model with a new label
- Adversarial neural network
Theory is explored in: https://thesis.unipd.it/handle/20.500.12608/33543