Attribution monitoring flags that the influence of a single input, 'territory_code', has risen markedly over the last seven days compared with its baseline. Which interpretation is most reasonable?
Show answer and explanation
Answer: D
Attribution drift measures how much each feature contributes to predictions over time. A rise for one feature means the model is relying on it more than during the baseline window. That may reflect real market changes or a data issue, so the team should investigate whether this reliance is appropriate before deciding on retraining.
Why the other options are wrong
- A. The monitor is not necessarily faulty. A genuine shift in how the model weights the input is exactly what this signal is designed to reveal, so the alert should be investigated first.
- B. A change in encoding would usually appear as skew in the input distribution. Attribution drift describes how predictions depend on inputs, which is a separate signal.
- C. Attribution drift is a reason to investigate, not automatic proof that retraining is needed. The cause could be legitimate or a data problem, so jumping to retraining skips diagnosis.