I have different features in my train and test set for Random Forest

Should I choose only the important features from the train set and use that for predictions or create columns with 0 values for those features not included?

training set 6160 features test set 4000 features