By: Žan Mervič, Sep 19, 2023
In this blog, we confront the common misconception that merely removing a protected attribute from a dataset eliminates bias in model predictions. Our case study reveals that models trained without these attributes still produce biased results. This is due to feature correlations that indirectly capture the protected information. Our conclusion? You cannot sidestep the need for specialized fairness algorithms.
By: Žan Mervič, Sep 19, 2023
Diving deeper into the Orange fairness Reweighing widget, we explore its use as a preprocessor for models. Discover the new widgets and fairness scoring metrics; all illustrated using the German credit dataset, supplemented with visual insights through box plots.
By: Žan Mervič, Sep 19, 2023
Building on our exploration of the Orange fairness addon, this blog delves into the Reweighing widget. By adjusting weights for dataset instances, the widget addresses bias, focusing on underrepresented groups. Using the Compas dataset as an example, we demonstrate how bias decreases post-reweighting, presenting visual insights into the distribution of adjusted weights and their impact on dataset fairness.
By: Žan Mervič, Sep 19, 2023
In this blog, we delve into the Equal Odds Postprocessing widget, a tool designed to enhance fairness in machine learning models. We break down how the algorithm works by modifying predictions to meet Equalized Odds criteria. Using a real-world example with the German credit dataset, we demonstrate its efficacy in improving fairness metrics while marginally affecting accuracy.
By: Žan Mervič, Sep 19, 2023
This blog post focuses on the Adversarial Debiasing model in Orange, a tool for enhancing fairness in your machine learning algorithms. We will walk through how to use it and explain the trade-offs that come with using fairness algorithms.
By: Žan Mervič, Sep 18, 2023
In an era where AI drives decisions impacting real lives, fairness in machine learning is paramount. Take the `Adult` dataset, which shows discrepancies in salary predictions based on gender. Addressing such concerns, Orange introduces a fairness add-on. Using new widgets, users can identify and mitigate biases in their datasets or model predictions.
