A Simulation Study of Indirect Prognostic Classification

Abstract

Prognostic classification refers to identifying the class to which an object is likely to belong in the future. In many situations the classes are defined indirectly via some intermediate variables whose values are not known at present and only become known in the future. An example of this framework is provided in the classification of degrees at our university. The degree class is determined from continuous assessment work and examination scores, using deterministic rules. The standard classification approach (the ‘direct’ approach) would predict the degree grade directly from initial information, such as age, high school performance and so on. An alternative approach, previously unexplored, is to predict the intermediate variables and then apply the classifying rule set. We term this ‘indirect prognostic classification’, and describe the method in Section 2.