QuestionQ72

VerifiedData Preparation for Machine Learning (ML)

An ML engineer is preparing data for binary-classification modeling. The ML engineer runs an Amazon SageMaker Clarify processing job, and the bias report shows a difference in proportions of labels (DPL) score of 0.4. The ML engineer wants label occurrences to be equal across classes and decides to use SageMaker Data Wrangler to balance the data.

What should the ML engineer do to most of the data to meet these requirements?

Explanation

A DPL value near zero indicates more equal proportions of positive outcomes between facets. Random undersampling removes samples from the overrepresented category to achieve the desired class proportion, so it balances the data while reducing the DPL score. AWS: Difference in Proportions of Labels (DPL) AWS: Transform Data—Balance Data

Learn more

Community Discussion

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