A healthcare organization uses an AI model to analyze patient data and provide diagnostic recommendations. Which of the following MOST effectively detects data drift related to the model's…

Question

Options

• AComparing incoming patient data distributions with the training data set
• BApplying overrides to allow healthcare professionals to correct the AI model's recommendations
• CConducting periodic model retraining to ensure alignment with updated patient data
• DUsing adversarial testing to simulate scenarios that stress test the model's predictions

• A
  87% (27)
• B
  3% (1)
• C
  3% (1)
• D
  6% (2)

Explanation

Data drift occurs when the statistical properties of incoming real-world data shift away from the distribution of the original training data. Comparing incoming patient data distributions against the training dataset (A) directly identifies this divergence, making it the most effective detection method. Allowing healthcare professionals to override recommendations (B) corrects errors but does not detect drift. Periodic retraining (C) addresses drift after it is known but is not a detection mechanism. Adversarial testing (D) stress-tests model robustness but does not monitor for distributional changes in production data.

No community discussion yet for this question.