Question 650
At an enterprise engineering an AI prediction system, an AI project leader faces this situation: A churn classifier reports unusually high accuracy because future cancellation dates accidentally entered features. The training process must remain consistent with intended deployment constraints. Which plan change would best correct the situation?
Explanation
Correct Answer
In an enterprise engineering an AI prediction system, the most direct response is: Audit feature availability timestamps and rebuild splits using valid pre-decision information. This addresses: Data leakage can create deceptively strong test results that collapse when the model is deployed. Its acceptance evidence is: A leakage test excludes future-known features and reports realistic holdout quality. The project should also remember: Do not accept impressive metrics without checking how the training examples were constructed.
Incorrect Answers
Create a training schedule with checkpoint frequency and retry behavior suited to the workload. The appropriate correction for the reported case is: Audit feature availability timestamps and rebuild splits using valid pre-decision information.
Freeze approved artifacts, parameters and code revisions before release testing. This alternative is not the failing dependency described in the question.
Add robustness cases covering degraded images, unexpected formats and incomplete documents. Its verification is distinct from the reported condition: A churn classifier reports unusually high accuracy because future cancellation dates accidentally entered features. The training process must remain consistent with intended deployment constraints.