Happiness as a Measure of Fairness
Georg Pichler, Marco Romanelli, Pablo Piantanida
TL;DR
The paper introduces a happiness-based fairness framework for post-processing soft classifiers, where a vector-valued happiness function $\boldsymbol{\eta}$ captures group utility from outcomes. Fairness is defined by an $\varepsilon$-bounded difference in average happiness across groups, and the optimal post-processing rule $p_{\tilde{Y}|\hat{Y}Z}$ is found by solving a linear program that minimizes misclassification loss. The framework unifies several standard fairness notions (statistical parity, overall accuracy, equalized odds) as special cases via appropriate choices of $\boldsymbol{\eta}$, and is demonstrated through synthetic and real-world case studies (including loan approval and the Adult dataset) to achieve favorable fairness-utility trade-offs with minimal accuracy loss. The method is scalable, requiring only a linear program and enabling application to resource allocation problems where happiness or utility, not just accuracy, matters for fairness.
Abstract
In this paper, we propose a novel fairness framework grounded in the concept of happiness, a measure of the utility each group gains fromdecisionoutcomes. Bycapturingfairness through this intuitive lens, we not only offer a more human-centered approach, but also one that is mathematically rigorous: In order to compute the optimal, fair post-processing strategy, only a linear program needs to be solved. This makes our method both efficient and scalable with existing optimization tools. Furthermore, it unifies and extends several well-known fairness definitions, and our empirical results highlight its practical strengths across diverse scenarios.