Quiz: Mitigation: Pre Processing, In Processing and Post Processing

Pre processing mitigation aims to reduce unfair outcomes by addressing bias before data enters machine learning models through techniques such as stratified sampling or oversampling.

In recruitment systems, discriminatory features such as candidate names or addresses should be identified early in the data pipeline to prevent unintentional bias.

In processing mitigation, equalized odds constraints are applied to ensure similar false positive rates across different demographic groups during model training.

Post processing mitigation involves adjusting model outputs after predictions have been generated by modifying decision thresholds or reweighting predictions to achieve fairness.

The computational cost of adversarial debiasing techniques increases but provides measurable improvements in fairness metrics in criminal justice applications.