Masked Hiring– Masking of Personal Identifiable Information (PII) during screening to reduce unconscious bias.
Model Bias Analysis-Disparate impact analysis using the 4/5 rule on statistical parity difference is used to detect bias in models across sensitive attributes/ groups. Propensity score matching is also used to compare the predicted outcome differences.
Regular Model Review & Retraining- We review our models regularly (model decay, data & concept drift) and check our data on potential biases before using it to retrain our models. No sensitive attribute is used as part of model training/retraining.