École Nationale Supérieure de
Techniques Avancées

Résumé

his report explores the impact of sampling bias on machine learning models and evaluates the effectiveness of two bias mitigation techniques—Reweighing and Equalized Odds—across two datasets: Adult and COMPAS. Sampling bias, including Underrepresentation Bias (URB) and Sample Size Bias (SSB), is analyzed and treated. Bias metrics are calculated before and after applying the mitigation techniques to assess their effectiveness in promoting fairness. The results highlight the challenges and trade-offs involved in addressing bias within machine learning systems, providing valuable insights for developing more equitable models.