Debiaser for Multiple Variables to enhance fairness in classification tasks

Giordano d'Aloisio, Andrea D'Angelo, Antinisca Di Marco, Giovanni Stilo

Abstract: Highlights•We highlight the strengths and weaknesses of established bias mitigation methods.•We revise our work by generalizing the instance generation and removal strategies.•We evaluate the effectiveness of different instance-generation strategies in DEMV.•We evaluate DEMV by considering several datasets, methods and sensitive variables.•We observe how DEMV overcomes the baselines in almost all the experiments.