TreeWorks: Advances in Scalable Decision Trees

Abstract

Decision trees are hierarchical, sequential classification structures that recursively partition the set of observations (data) and are used to represent rules underlying the observations. This article describes the development of TreeWorks, a tool that enhances existing decision tree theory and overcomes some of the common limitations such as scalability and the ability to handle large databases. We present a heuristic that allows TreeWorks to cope with observation sets that contain several distinct values of categorical data, as well as the ability to handle very large datasets by overcoming issues with computer main memory. Furthermore, our tool incorporates a number of useful features such as the ability to move data across terminal nodes, allowing for the construction of trees combining statistical accuracy with expert opinion. Finally, we discuss ways that decision trees can be combined with Operational Research health care models, for more effective and efficient planning and management of health care processes.