MLA-C01 Question 95
Select 4You are developing a machine learning model to classify images of animals into different species. During the initial analysis, you notice a significant class imbalance, with most of the data belonging to common species and very few samples for rare species. What strategies could you use to address this issue effectively?
- A
Use synthetic data generation techniques like GANs (Generative Adversarial Networks) to augment the dataset with more images of rare species.
- B
Apply oversampling by duplicating images of rare species to balance the dataset.
- C
Apply undersampling by reducing the number of images of common species to match the rare species.
- D
Remove all data for rare species to simplify the model and reduce complexity.
- E
Use weighted loss functions to penalize misclassifications of rare species more heavily during training.
Show answer and explanation
Correct answers: A, B, C, E
Explanation
Class imbalance in datasets can negatively impact machine learning model performance, especially for underrepresented classes. Strategies such as synthetic data generation, oversampling, undersampling, and using weighted loss functions are effective ways to address this issue. Removing rare class data is not a valid solution, as it defeats the purpose of solving the imbalance problem.
- A. Correct.
Using synthetic data generation techniques like GANs is a common approach to augment datasets with additional realistic samples and can help balance the dataset effectively.
- B. Correct.
Oversampling by duplicating rare class samples is a straightforward and widely used method to address class imbalance without losing valuable data.
- C. Correct.
Undersampling the majority class helps balance the dataset, although it risks losing information from the majority class and should be applied cautiously.
- D. Incorrect.
Removing data for rare species is counterproductive as it eliminates the very classes you are trying to improve predictions for.
- E. Correct.
Using weighted loss functions is an effective way to address class imbalance during training without altering the dataset itself, by assigning higher penalty weights to rare classes.