4.2. Naive Bayesian learner
While the attributes of the problem do not adhere to
the strict independence assumption of the naive Bayesian
learner we would expect there to be a reasonable match
and thus for this learner to perform relatively well. This is
reXected in that for non-overlapping training and test
data sets on the general model this learner came second
overall with a classiWcation error of 61.19%. Interestingly
on the expert chosen data the naive Bayesian learner only
came in Wfth best with a classiWcation error of 64.26%.