Transplanting Binary Decision Trees

Abstract

In this paper, we describe a means of compiling binary decision trees as generated by the C4.5 binary decision tree classifier into high-performance, reusable, stand-alone, run-time classifiers. We demonstrate the memory savings and run time characteristics of a compiled tree as compared to the traditional use of a C4.5 runtime. We demonstrate 100% correctness over every input we have available for testing as compared to our own enhanced version of the classic C4.5 run-time classification routine, consultr. In addition, this work provides a framework for comparing decision tree classifiers to more in vogue classifiers such as support vector machines as demonstrated within.