An AI model used for predicting hospital no-shows achieves 78% accuracy overall, but the AI risk team discovers it has a false negative rate of 35% for elderly patients. What action is MOST appropriate?
Select an answer to reveal the explanation.
Short Explanation
You can't ignore a 35% false negative rate for elderly patients just because overall accuracy looks fine. That gap may be age discrimination, so you need to investigate and see if retraining with stratified sampling fixes it.
Full Explanation
The correct action is to investigate whether the false negative rate disparity violates age discrimination regulations and assess whether model retraining with stratified sampling is required. A significantly higher false negative rate for elderly patients suggests systematic underperformance for this protected group, which may constitute age discrimination under applicable laws. The investigation should determine causation and whether retraining with stratified sampling can reduce the disparity. Accepting the model because overall accuracy meets the 75% benchmark ignores a material subgroup performance gap. Excluding elderly patients and applying manual scheduling may itself be discriminatory. Increasing the global decision threshold may worsen performance for other groups.