Boosting of Fuzzy Rules with Low Quality Data
Ana M. Palacios, Luciano Sanchez and Ines Couso
The task of modeling and reasoning about real-world problems often involves analyzing over-constrained representations, where not all constraints of a problem can be simultaneously satisfied. The need to analyze over-constrained (or unsatisfiable) problems occurs in many settings, including data and knowledge bases, artificial intelligence, applied formal methods, operations research and description logics. In most cases, the problem to solve is related with some form of minimal unsatisfiability, i.e. an irreducible set of constraints that explains unsatisfiability. This paper provides an overview of algorithms for computing minimally unsatisfiable subformulas, and conducts an experimental evaluation of these algorithms. In addition, the paper briefly overviews computational problems related with minimal unsatisfiability in Boolean logic, practical applications of minimal unsatisfiability, and extensions of minimal unsatisfiability to other domains.
Keywords: Genetic Fuzzy Systems, Low Quality Data, Boosting, Genetic Algorithms, Fuzzy Rule-based Classifiers, Vague Data