Addressing Fairness, Bias and Class Imbalance in Machine Learning: the FBI-loss

Resilience to class imbalance and confounding biases, together with the\nassurance of fairness guarantees are highly desirable properties of autonomous\ndecision-making systems with real-life impact. Many different targeted\nsolutions have been proposed to address separately these three problems,\nhowever a unifying perspective seems to be missing. With this work, we provide\na general formalization, showing that they are different expressions of\nunbalance. Following this intuition, we formulate a unified loss correction to\naddress issues related to Fairness, Biases and Imbalances (FBI-loss). The\ncorrection capabilities of the proposed approach are assessed on three\nreal-world benchmarks, each associated to one of the issues under\nconsideration, and on a family of synthetic data in order to better investigate\nthe effectiveness of our loss on tasks with different complexities. The\nempirical results highlight that the flexible formulation of the FBI-loss leads\nalso to competitive performances with respect to literature solutions\nspecialised for the single problems.\n