QuestionQ135

VerifiedPreparing and using data for analysis

You are creating a model to predict whether it will rain on a particular day. You have thousands of input features and want to determine whether removing some can improve training speed with minimal effect on model accuracy. What can you do?

Explanation

Highly co-dependent features often encode overlapping information. Combining such a group into one representative feature reduces input dimensionality and training cost while retaining most of the shared predictive signal. Features highly correlated with the output label are ordinarily valuable predictors, whereas arbitrary averaging or dropping features solely because of a missing-value threshold can discard useful information.

Learn more

Community Discussion

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