Which of the following BEST helps in detecting AI model drift?

Detecting model drift requires a point of comparison and real-time visibility. The AAIA™ manual identifies the best practice as " Establishing a performance baseline " (using metrics like accuracy or F1-score from the initial validation phase) and then " Implementing continuous monitoring " to track those metrics as the model processes live production data. Any significant deviation from the baseline serves as an early warning that the model ' s environment has shifted and retraining is necessary. Periodic reviews (Option A) are helpful but may not detect drift quickly enough to prevent erroneous decisions in dynamic environments.