QuestionQ47

VerifiedData Preparation for Machine Learning (ML)

An ML engineer is fine-tuning an image-classification model that predicts one of two available classes poorly. Analysis shows that the images in the poorly predicted class make up an extremely small portion of the entire training dataset.

The ML engineer must improve the model’s performance.

Which solution meets this requirement?

Explanation

For a highly imbalanced binary image-classification dataset, F1 is preferable to accuracy because it reflects both precision and recall rather than allowing the majority class to dominate the metric. Image augmentation can expand the minority class with valid transformed image examples, improving representation during training. SMOTE interpolates feature vectors and is not generally appropriate for generating raw image samples.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!