Distance-Based Fuzzy-Rough Sets and Their Application to the Classification Problem

We propose distance-based fuzzy-rough sets (DBFR) that rely on distance functions for the granulation of the underlying universe. The classification problem is investigated from a fuzzy perspective and cast as a concept approximation problem. DBFRs are employed to facilitate the approximation process. The geometrical nature of approximation emerging due to the use of distance functions is investigated. Naive classifiers based on DBFRs are proposed and experimentally evaluated on benchmark datasets.