Partial Tree-Edit Distance: A Solution to the Default Class Problem in Pattern-Based Tree Classification

Pattern-based tree classifiers are capable of producing high quality results, however, they are prone to the problem of the default class overuse. In this paper, we propose a measure designed to address this issue, called partial tree-edit distance (PTED), which allows for assessing the degree of containment of one tree in another. Furthermore, we propose an algorithm which calculates the measure and perform an experiment involving pattern-based classification to illustrate its usefulness. The results show that incorporating PTED into the classification scheme allowed us to significantly improve the accuracy on the tested datasets.