In deep learning tasks, when encountering data imbalance problems, which of the following methods can we use to solve the problem?

Question

Options

• Abatch deletion
• BRandom oversampling
• CSynthetic sampling
• DRandom undersampling

• A
  17% (6)
• B
  83% (29)

Explanation

Options B, C, and D are all valid strategies for addressing class imbalance: Random oversampling (B) duplicates minority-class samples to balance the dataset, Synthetic sampling (C) - such as SMOTE - generates new artificial minority-class examples by interpolating between existing ones, and Random undersampling (D) removes majority-class samples to reduce the imbalance. Each approach adjusts the training distribution so the model doesn't develop a bias toward the dominant class.

Option A, batch deletion, is not a recognized technique for handling imbalance - deleting batches arbitrarily would discard data without any principled balancing logic, and would more likely degrade model performance than improve it.

Memory tip: Think "add, create, or remove" - you can add copies (oversampling), create new samples (synthetic), or remove excess (undersampling). Any other operation (like random batch deletion) doesn't fit that logic and is a distractor.

No community discussion yet for this question.