A comparative analysis of pruning strategies for fuzzy decision trees

Decision trees are powerful models which can be applied to classification tasks. Fuzzy decision trees unite the advantages of the classic decision trees, which produce simple models with high interpretability, competitive accuracy, and a graphical representation, as well as the advantages of fuzzy systems, which include the capability of dealing with imprecision and uncertainty in data. A key issue for the induction of decision trees is the pruning process usually applied to avoid overfitting of the induced models. The two basic types of pruning applied to decision trees are prepruning and postpruning. In this paper, we experimentally evaluate these two types of pruning, as well as no pruning, using a fuzzy decision tree, based on the C4.5 classic decision tree algorithm, namely FuzzyDT. The obtained results are compared with those of C4.5 in terms of the accuracy and interpretability of the induced models generated for 15 datasets. The postpruning strategy shows better accuracy rates than the other approaches, producing more interpretable models as well.