Bias detection and mitigation is the correct capability because it directly measures and helps remediate disparate outcomes across sensitive attributes and it enables the team to validate equitable treatment and meet anti discrimination requirements.

SageMaker Clarify provides bias analysis during training and inference and it can compute fairness metrics and produce reports that help the team identify and reduce discriminatory outcomes. Using bias detection and mitigation lets the data science group test model behavior across user segments and implement corrective actions before deployment.

Model compression techniques focus on reducing model size and improving latency which helps deployment but they do not detect or correct unfair predictions across demographic groups.

Amazon SageMaker Automatic Model Tuning searches for hyperparameters to improve model performance and it does not evaluate or mitigate bias across protected classes.

Cross-validation improves confidence that the model will generalize and it reduces overfitting but it does not surface inequities among protected classes and therefore does not ensure fairness.

Full AWS Practitioner Certification Question