R - naive bayes gives different result after binarisation

Question

My datas are spam. I have 4601 lines containing frequencies of string and the last column type has two values possible: spam or nonspam.

I have made