If you reopen your J48 learner, you should find a column marked in red on the right of the panel, it could give you a hint on variables that give you trouble. It seems to me that J48 is a C4.5 variant, so it is made to cope with nominal values or strings. You can send me a sample of your entry data set just before the J48 learner in knime format precising the variable you want to predict. I'll check.
By the way I noticed that the 3.7 Weka / knime embedded version has some strange behaviors comparing to the previous 3.6 (proof in attachment). So give it a try.