SE-Trees Outperform Decision Trees in Noisy Domains

All Contents

As a classifier, a Set Enumeration (SE) tree can be viewed as a generalization of decision trees. At the cost of a higher complexity, a single SE-tree encapsulates many alternative decision tree structures. An SE-tree enjoys several advantages over decision trees: it allows for domain-based user-specified bias; it supports a flexible tradeoff between the resources allocated to learning and the resulting accuracy; and it can combine knowledge induced from examples with other knowledge sources. We show that SE-trees enjoy a particular advantage over simple decision trees in noisy domains. This advantage manifests itself both in terms of accuracy, and in terms of consistency.