Evaluation of Classification Trees - Data Mining with Decision Trees: Theory and Applications - page 60

of its “parents” obviating the need to try each one and simplifying the
knowledge acquisition task. Even more ambitiously, there is hope that this
combination of paradigms might produce synergistic effects (for instance,
by allowing different types of frontiers between classes in different regions
of the example space), leading to levels of accuracy that neither atomic
approach by itself would be able to achieve.
Unfortunately, this approach has often been used with only moderate
success. Although it is true that in some industrial applications (like in
the case of demand planning) this strategy proved to boost the error
performance, in many other cases the resulting algorithms are prone to
be cumbersome, and often achieve an error that lies between those of their
parents, instead of matching the lowest.
The dilemma of which method to choose becomes even greater, if other
factors such as comprehensibility are taken into consideration. For instance,
for a specific domain, a neural network may outperform decision trees in
accuracy. However, from the comprehensibility aspect, decision trees are
considered better. In other words, even if the researcher knows that neural
network is more accurate, he still has a dilemma regarding which method
to use.