Aequitas is an algorithm that detects biasedness in machine learning datasets and automatically corrects it using a directed search method.
We improved the algorithm to account for nonbinary sensitive features. Additionally, the algorithm now allows for more than one sensitive features within an input dataset.
Check out our GitHubWe refactored the Aequitas codebase to be more reusable, and published it as a Pypi package. It is called 'Phemus' because 'Aequitas' was already taken.
Download Phemus packageWe developed a user-friendly website built with React frontend and Django backend, using Google Drive and EmailJS APIs. Users can upload their dataset and get retraining dataset and/or the improved model outputted by Aequitas.
Visit AequitasWeb