Missing values are harmful to machine learning models! not sure how many % is true on this..

Missing values are harmful to machine learning models! not sure how many % is true on this..

But one thing is sure from a beginner perspective, if we handle missing values then the model accuracy will be improved..

With that note I’ve started into spend some more time with N/A values,

1. When to handle missing values, before splitting the data set or after splitting the data set. what will be the diff and in which may my accuracy will be improved
2. Upon searching online I’ve found, the missing values will be falls under 3 categories MCAR, MAR, MNAR.
3. But upon these 3 categories, I’m searching my column in data set contains missing values and in which category does it lies, and how to find that?