fairmodels: a Flexible Tool for Bias Detection, Visualization, and Mitigation in Binary Classification Models

Machine learning decision systems are becoming omnipresent in our lives. From dating apps to rating loan seekers, algorithms affect both well-being and future. Typically, however, these not infallible. Moreover, complex predictive models eager learn social biases present historical data that may increase discrimination. If we want create responsibly, need tools for in-depth validation of also from potential This article introduces an R package fairmodels helps validate fairness eliminate bias binary classification quickly flexibly. The offers a model-agnostic approach detection, visualization, mitigation. implemented functions metrics enable model different perspectives. In addition, the includes series methods mitigation aim diminish discrimination model. is designed examine single facilitate comparisons between multiple models.