MLA-C01 Question #91: Real Exam Question with Answer & Explanation

Question

An ML engineer notices class imbalance in an image classification training job. What should the ML engineer do to resolve this issue?

Options

• AReduce the size of the dataset.
• BTransform some of the images in the dataset.
• CApply random oversampling on the dataset.
• DApply random data splitting on the dataset.

Explanation

Random oversampling (C) directly addresses class imbalance by duplicating samples from underrepresented classes, giving the model more balanced exposure across all categories without discarding any data. Reducing dataset size (A) worsens the problem - you'd lose even more minority-class examples. Transforming images (B) is data augmentation, which can help generalization but doesn't fix the underlying imbalance ratio between classes. Random data splitting (D) is a partitioning strategy (train/val/test) that has no effect on class distribution within those splits.

Memory tip: Think of oversampling as "copying the underdog" - you're giving minority classes more "votes" in training so the model stops ignoring them. The key word is over (adding more), not under (removing majority samples), though both are valid techniques.

No community discussion yet for this question.