classification Trees

#1
Hello
I'm doing a research about the application of classification trees on drop out education.
I was confused about deciding what results should be presented, I used three growing methods : CHAID , EXHAUSTIVE CHAID and CRT.
the evaluation by using re-substitution and cross validation risks and overall correct percentage and that was in favour of Exhaustive CHAID method, but the results of the generated tree was not logic and don;t make sense at all.
on the other hand the generated tree of CRT and CHAID methods was more informative and have meaningful interpretation.
note that the re-substitution and cross validation risks and overall correct percentage are approximately equal for all growing methods.
my question is should I take the most interpretable generated tree at the expense of re-substitution and cross validation risks.