Case context: A machine learning engineer reviews errors in a bank card fraud detector and finds that it misses a group of transactions with an unusual combination of merchant category and time-of-day patterns. To address that specific weakness, the engineer adds new engineered input features that highlight those patterns. The training error falls, but validation performance now shows a noticeable rise in variance.

Question: According to the standard approach for handling this tradeoff after improving bias with new features, what should the engineer do next to address the increased variance?

Sample answer: The engineer should apply regularization, which is the usual next step for reducing the variance that can appear after adding features intended to lower avoidable bias.

Key points:

• The added features were designed to reduce avoidable bias in a targeted error category.
• The side effect observed was an increase in variance.
• Regularization is the appropriate technique for counteracting that variance increase.

Rubric: The answer must correctly identify regularization as the needed intervention to address the rise in variance.