An AI model for predicting equipment failure in a nuclear facility is flagged for review because its training data was collected over only six months, during summer operations. The facility also operates during winter with significantly different temperature and load conditions. What risk does this introduce?
Select an answer to reveal the explanation.
Short Explanation
You've got a temporal bias problem here. The model learned from summer-only data, so it'll perform better in summer and worse in winter when temperature, load, and failure patterns shift. That's a sampling bias that leaves winter conditions underrepresented.
Full Explanation
The correct risk is temporal bias that will cause the model to perform better in summer than winter due to underrepresentation of winter conditions. Because the model trained only on summer data, it learned patterns specific to summer operating conditions and will encounter distributional shift in winter, when temperature, load profiles, and failure mechanisms may differ substantially. This is a form of temporal/contextual sampling bias that creates systematic underperformance in winter. The data leakage concern involving winter operational data influencing training through indirect channels is a different issue involving inadvertent use of test data information during training. The model drift concern related to upgrading sensor equipment to higher-accuracy models relates to post-deployment changes, not training data limitations. The label imbalance concern because equipment failures are rare events underrepresented in any six-month period is a valid general concern for rare events but is not the specific risk identified by the summer-only training window.