Let's say we have supervised learning problem with a dataset consisting of 100 features and 100k labeled data points (numbers don't matter).

Why is feature selection and feature extraction such an import task?

Intuitively, throwing away features aka information seems a bad idea. Furthermore I thought that it's our model's job to find out which features are important and at which scale. So we should we do it manually? Why can't our model do it itself?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.