Classification Tree with Hybrid Splitting Mechanism

Abstract

Binary Decision Trees (BDT), such as CART and C4.5, is one of the most commonly used classifiers, where each node of the tree can be split into a new layer with two child nodes. An alternative classification tree, Multi-Layer Classifier (MLC) has been proposed to take advantage of the underlying data structure. The MLC splits the parent node into 1 or 2 classified child node and an unclassified child node at each layer. Unlike BDT, MLC only allows the unclassified child node to be split into a new layer until a stop criterion is reached. In this study, an algorithm that switches between the two splitting mechanisms in the tree construction process is proposed. In the experiments of real cases, the proposed classifier has better performance than BDT and MLC in terms of the out-of-sample accuracy and Youden’s Index.