You're training a supervised learning model and have many candidate features, some redundant, some noisy, and some only weakly useful.
What is your approach to feature selection in machine learning?
