Skip to Main Content
In many disciplines, such as social and behavioral sciences, we often have to do ordinal classification by assigning objects to ordinal classes. The fundamental objective of ordinal classification is to create an ordering in the universe of discourse. As such, a decision tree for ordinal classification should aim at producing an ordering which is most consistent with the implicit ordering in the input data. Ordinal classification problems are often dealt with by treating ordinal classes as nominal classes, or by representing the classes as values on a quantitative scale. Such approaches may not lead to the most desirable results since the methods do not fit the type of data, viz. ordinal data, concerned. In this paper, we propose a new measure for assessing the quality of output from an ordinal classification approach. We also propose an induction method to generate an ordinal decision tree for ordinal classification based on this quality perspective. We demonstrate the advantage of our method using results from a set of experiments.