Feature Selection and Disambiguation in Learning from Fuzzy Labels Using Rough Sets

In this article, we study the setting of learning from fuzzy labels, a generalization of supervised learning in which instances are assumed to be labeled with a fuzzy set, interpreted as an epistemic possibility distribution. We tackle the problem of feature selection in such task, in the context of rough set theory (RST). More specifically, we consider the problem of RST-based feature selection as a means for data disambiguation: that is, retrieving the most plausible precise instantiation of the imprecise training data. We define generalizations of decision tables and reducts, using tools from generalized information theory and belief function theory. We study the computational complexity and theoretical properties of the associated computational problems.