In this paper, a variant of C4.5 decision tree algorithm named NeC4.5 is proposed, which utilizes neural network ensemble to preprocess the training data for decision tree
induction. Such an algorithm can work well because the original training set may contain much noise and may not capture the whole target distribution. Since its learning results can be more accurate than that of C4.5 while the reasoning process remains explicitly explainable, NeC4.5 provides a good choice for tasks where both the generalization ability and the comprehensibility are concerned