You are training a supervised learning model and notice that several input features have missing values. You need a practical way to prepare the data without distorting the signal or introducing leakage.
How would you handle a dataset with missing values?