Abstract:The reason behind the unfair outcomes of AI is often rooted in biased datasets. Therefore, this work presents a framework for addressing fairness by debiasing datasets containing a (non-)binary protected attribute. The framework proposes a combinatorial optimization problem where heuristics such as genetic algorithms can be used to solve for the stated fairness objectives. The framework addresses this by finding a data subset that minimizes a certain discrimination measure. Depending on a user-defined setting, the framework enables different use cases, such as data removal, the addition of synthetic data, or exclusive use of synthetic data. The exclusive use of synthetic data in particular enhances the framework's ability to preserve privacy while optimizing for fairness. In a comprehensive evaluation, we demonstrate that under our framework, genetic algorithms can effectively yield fairer datasets compared to the original data. In contrast to prior work, the framework exhibits a high degree of flexibility as it is metric- and task-agnostic, can be applied to both binary or non-binary protected attributes, and demonstrates efficient runtime.
Abstract:Motivated by the recital (67) of the current corrigendum of the AI Act in the European Union, we propose and present measures and mitigation strategies for discrimination in tabular datasets. We specifically focus on datasets that contain multiple protected attributes, such as nationality, age, and sex. This makes measuring and mitigating bias more challenging, as many existing methods are designed for a single protected attribute. This paper comes with a twofold contribution: Firstly, new discrimination measures are introduced. These measures are categorized in our framework along with existing ones, guiding researchers and practitioners in choosing the right measure to assess the fairness of the underlying dataset. Secondly, a novel application of an existing bias mitigation method, FairDo, is presented. We show that this strategy can mitigate any type of discrimination, including intersectional discrimination, by transforming the dataset. By conducting experiments on real-world datasets (Adult, Bank, Compas), we demonstrate that de-biasing datasets with multiple protected attributes is achievable. Further, the transformed fair datasets do not compromise any of the tested machine learning models' performances significantly when trained on these datasets compared to the original datasets. Discrimination was reduced by up to 83% in our experimentation. For most experiments, the disparity between protected groups was reduced by at least 7% and 27% on average. Generally, the findings show that the mitigation strategy used is effective, and this study contributes to the ongoing discussion on the implementation of the European Union's AI Act.
Abstract:In this paper, we deal with bias mitigation techniques that remove specific data points from the training set to aim for a fair representation of the population in that set. Machine learning models are trained on these pre-processed datasets, and their predictions are expected to be fair. However, such approaches may exclude relevant data, making the attained subsets less trustworthy for further usage. To enhance the trustworthiness of prior methods, we propose additional requirements and objectives that the subsets must fulfill in addition to fairness: (1) group coverage, and (2) minimal data loss. While removing entire groups may improve the measured fairness, this practice is very problematic as failing to represent every group cannot be considered fair. In our second concern, we advocate for the retention of data while minimizing discrimination. By introducing a multi-objective optimization problem that considers fairness and data loss, we propose a methodology to find Pareto-optimal solutions that balance these objectives. By identifying such solutions, users can make informed decisions about the trade-off between fairness and data quality and select the most suitable subset for their application.
Abstract:The literature for fairness-aware machine learning knows a plethora of different fairness notions. It is however wellknown, that it is impossible to satisfy all of them, as certain notions contradict each other. In this paper, we take a closer look at academic performance prediction (APP) systems and try to distil which fairness notions suit this task most. For this, we scan recent literature proposing guidelines as to which fairness notion to use and apply these guidelines onto APP. Our findings suggest equalised odds as most suitable notion for APP, based on APP's WYSIWYG worldview as well as potential long-term improvements for the population.