Model tree pruning

• DOI: 10.1007/s13042-019-00930-9
• Publication Date: 2019-02-01T20:47:21Z
• Authors: Xinlei Zhou; Dasen Yan
• Keywords: 0202 electrical engineering, electronic engineering, information engineering; 02 engineering and technology

A model tree is a decision tree in which a specified model, such as a linear regression or naive Bayes model, is built on part of the leaf nodes. Compared with the typical decision tree in which every leaf node is assigned a class label, a model tree has several advantages: the flexibility to handle mixed attributes, a simplified tree structure, and a good potential for processing big data. This paper investigates a model tree in which the ELM model is applied to some leaf nodes of the tree and compares two fundamental strategies for generating model trees in terms of training complexity and generalization ability, namely, prepruning and postpruning. The experimental results and algorithmic analysis show that, with respect to the ELM model tree, postpruning achieves better performance than does prepruning, which has previously been universally regarded as one of the most popular decision tree generation strategies.