Is there any best practice to deal with missing value before applying machine learning algorithm

We often get data with missing values. Records having missing value are removed or replaced with mean or mode. Is there any best practice related to dealing with missing value