Simultaneous Fuzzy Rough Prototype Selection and Evolutionary Feature Selection

Date of Presentation: 
Wednesday, August 6, 2014
Faculty Sponsor: 
Martine De Cock
Quarter: 
2014 Summer
Research Focus: 

Speaker: Nele Verbiest
Title: Simultaneous Fuzzy Rough Prototype Selection and Evolutionary Feature Selection

Faculty Sponsor: Martine De Cock

Abstract: Fuzzy rough set theory, the hybridization between rough set theory and fuzzy set theory, is a mathematical framework that allows to model inconsistencies in real-valued data. As a result, fuzzy rough set theory is a perfect tool to carry out prototype selection, the process where objects are removed from the data before using it for nearest neighbor classification. In this talk I will present Fuzzy Rough Prototype Selection (FRPS), a prototype selection technique based on fuzzy rough set theory. Secondly, I will discuss evolutionary algorithms, search algorithms based on natural selection. This type of algorithms is very successful for feature selection, which removes irrelevant features from the data before using it for other data mining purposes. As both FRPS and evolutionary feature selection techniques are very successful on there own, it is worth trying to combine them. As the sequential application of feature selection and prototype selection might pose problems - the feature selection process should not be based on noisy instances and the prototype selection part should not rely on irrelevant features - we propose a technique that simultaneously applies both processes.

Bio: Nele Verbiest recently obtained her Ph. D. in Computer Science from Ghent University, Belgium. Her research interests are fuzzy rough set theory, evolutionary algorithms and instance selection.