Combination of Fuzzy Sets and Rough Sets for Machine Learning Purposes (Tutorial Lecture — Extended Abstract)

Fuzzy set theory is a popular AI tool designed to model and process vague information. Specifically, it is based on the idea that membership to a given concept, or logical truthhood of a given proposition, can be a matter of degree. On the other hand, rough set theory was proposed as a way to handle potentially inconsistent data inside information systems. In Pawlak's original proposal, this is achieved by providing a lower and upper approximation of a concept, using the equivalence classes of an indiscernibility relation as building blocks. Noting the highly complementary characteristics of fuzzy sets and rough sets, Dubois and Prade proposed the first working definition of a fuzzy rough set, and thus paved the way for a flourishing hybrid theory with numerous theoretical and practical advances. In this tutorial, we will explain how fuzzy rough sets may be successfully applied to a variety of machine learning problems. After a brief discussion of how the hybridization between fuzzy sets and rough sets may be achieved, including an extension based on ordered weighted average operators, we will focus on the following practical applications: 1. Fuzzy-rough nearest neighbor (FRNN) classification, along with its adaptations for imbalanced datasets and multi-label datasets 2. Fuzzy-rough feature selection (FRFS) 3. Fuzzy-rough instance selection (FRIS) and Fuzzy-rough prototype selection (FRPS) We will also demonstrate software implementations of all of these algorithms in the Python library fuzzy-rough-learn.

---

1. Main Track Invited Contributions

---

2. Opracowanie rekordu ze środków MEiN, umowa nr SONP/SP/546092/2022 w ramach programu "Społeczna odpowiedzialność nauki" - moduł: Popularyzacja nauki i promocja sportu (2024).