Fairness in Skin disease prediction

This project makes use of a medical dataset comprising images depicting skin disease in pedi- atric patients. These images have been previously augmented via data augmentation techniques and diffusion models to compensate for limited data availability. The main objective is to examine bias within the dataset, particularly considering its inherent bias towards Caucasian skin tones. Additionally, we plan to conduct experiments by injecting bias, for example by generating images of specific skin tones, to observe model’s behavior under varied conditions. We anticipate impli- cations related to fairness when testing the model on other skin color, and to this extent we aim to propose possible strategies to mitigate the phenomenon.