Partial classification in the belief function framework

Partial, or set-valued classification assigns instances to sets of classes, making it possible reduce the probability misclassification while still providing useful information. This paper reviews approaches partial based on Dempster–Shafer theory belief functions. To define utility predictions, we propose extend matrix using an Ordered Weighted Average operator, allowing us model decision maker’s attitude towards imprecision a single parameter. Various criteria are analyzed comprehensively. In particular, two main strategies distinguished: complete preorders among assignments, and assignments. Experiments with UCI simulated Gaussian data show superiority in terms average utility, as compared single-class assignment rejection.