Tree-based methods and decision trees
This chapter reviews a selection of tree-based methodologies that can be used for customer satisfaction evaluation. After a short introduction of some basic concepts, three popular methods – CART, CHAID and PARTY– are described and compared; particular attention is given to the treatment of missing values and to specific solutions suitable for dependent variables measured on an ordinal scale. Results obtained by analysing the ABC ACSS data set are illustrated and discussed. The chapter concludes by focusing on the main drawbacks of tree-based methods and discusses ways to overcome them.
15.1 An overview of tree-based methods and decision trees
15.1.1 The origins of tree-based methods
Tree-based methods are data-driven tools based on sequential procedures that recursively partition the data. They provide conceptually simple ways of understanding and summarizing the main features of the data; in particular, they exploit tree graphs to provide visual representations of the rules underlying a given data set. Automatic construction of such rules has been developed in the fields of statistics, engineering and decision theory, in the context of regression and classification analysis, pattern recognition and decision table programming, respectively. Recently, several contributions to this topic can be found in the literature on machine learning and neural networks. As a result, many real-world situations ...