Project: Fix Bias Using Threshold Optimization
We are already aware of the fact that changing the threshold can have a major impact on the performance of our model. As such, we plan to conduct an experiment with threshold selection in order to gain a more comprehensive understanding of this concept. Through this project, we hope to gain insight into the various effects of such threshold changes on the model.
We will load the predictions of the sample model and evaluate how we can influence them by selecting thresholds. We will use this knowledge to optimize it for achieving various fairness metrics without sacrificing too much of accuracy.