FARF: A Fair and Adaptive Random Forests Classifier

• DOI: 10.1007/978-3-030-75765-6_20
• Publication Date: 2021-05-07T09:08:54Z
• Authors: Wenbin Zhang; Albert Bifet; Xiangliang Zhang; Jeremy C. Weiss; Wolfgang Nejdl
• Keywords: Hyperparameter; Concept Drift; Ensemble Learning

As Artificial Intelligence (AI) is used in more applications, the need to consider and mitigate biases from learned models has followed. Most works developing fair learning algorithms focus on offline setting. However, many real-world applications data comes an online fashion needs be processed fly. Moreover, practical application, there a trade-off between accuracy fairness that accounted for, but current methods often have multiple hyperparameters with non-trivial interaction achieve fairness. In this paper, we propose flexible ensemble algorithm for decision-making challenging context of evolving settings. This algorithm, called FARF (Fair Adaptive Random Forests), based using component classifiers updating them according distribution, also accounts single alters fairness-accuracy balance. Experiments discriminated streams demonstrate utility FARF.