Question 40
Domain 3: Model DevelopmentA machine learning team finds that one class appears far less often than the others in its training set, and the model is performing poorly on that minority class. Which approach is the best way to mitigate this training-data imbalance?
Correct answer: B
Explanation
When class distributions are skewed, mitigation focuses on adjusting the training data so underrepresented classes are better represented during learning. — official.txt
Why each option is right or wrong
A. Remove additional examples from the minority class so all classes contribute equally to error rates
Mitigation aims to better support underrepresented classes, not reduce their presence further.
B. Rebalance the training data so the underrepresented class is better represented during model learning
The source material describes addressing skewed class distributions by using techniques so training data better supports model learning. In this scenario, the minority class appears far less often and performance on that class is poor, so rebalancing the training data to improve representation of the underrepresented class directly matches that purpose.
C. Leave the class distribution unchanged and rely on the model to infer minority patterns automatically
Skewed class distributions require mitigation so training data better supports model learning.
D. Increase the number of majority-class examples so the model sees the overall distribution more clearly
Adding more majority-class data increases skew instead of addressing class imbalance.