AHP-Gaussian To Enhance Model Selection Based On Multiple Fairness Criteria
Diego Minatel, Nícolas Roque dos Santos, Vinícius Ferreira, Mateus Modesto · 2024
The challenge of developing impartial models that minimize the propagation of unfair predictions is directly linked to optimizing multiple fairness concepts. Therefore, identifying which model best combines these concepts is essential for promoting fairness in machine learning. The field of Multi-Criteria Decision Analysis addresses similar issues by developing techniques for choosing the best alternative in complex problems. One standout method is AHP–Gaussian, which, through the Gaussian factor, defines the relevance of each criterion used in decision-making. This eliminates any human factor in weighing the criteria’s importance, making it an excellent alternative in the fairness-aware model selection task. To the extent of our knowledge, no study in the literature has proposed this approach before. This paper handles this gap and proposes applying AHP–Gaussian to select fairer models in classification tasks involving people. According to the results, AHP–Gaussian is more effective at selecting classifiers that balance predictive power and maximization of distinct fairness concepts than traditional multi-criteria methods.