The hierarchical, recursive tree construction methodology is simple and intuitively appealing. However, the simplicity of the methodology should not lead a practitioner to take a slack attitude towards using decision trees. Just as in the case of statistical methods or neural networks, building a successful tree classifier for an application requires a thorough understanding of the problem itself, and a deep knowledge of tree methodology.
This chapter attempted a multi-disciplinary survey of work in automatically constructing decision trees from data. We gave pointers to work in fields such as pattern recognition, statistics, machine learning, mathematical programming, neural networks etc. We attempted to provide a self-contained, concise description of the directions which decision tree work has taken over the years. Our larger goal is to help avoid some redundant, ad hoc effort, both from researchers and from system developers.