The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9822

Title:Genetic Algorithms for Tuning Classification Rules
Date: 1998
Abstract:One of the problems in Fuzzy Modelling, based on a combination of fuzzy production rules and fuzzy set definitions, is that in many cases the Fuzzy Rule Decision System (rules and sets) are not optimally adjusted to the data available to the system modelled. so a technique that allows a fast and accurate adjustment or tuning of such Fuzzy Models can improve greatly the reliability of the models.In this report we perform a comparison of several Evolutionary and Neighbourhood Search Techniques applied to the fine tuning of Membership Partitions used in Fuzzy Rule Decision Systems. Our goal is to check whether this task requires special heuristic methods that use information about the structure of the problem, or whether it is enough to use general purpose optimisers.We will also give several warnings about the extreme care that should be taken with the methodology of the experiments and point out several common mistakes that researchers may make.We will finally outline some potential extensions to the tuning task.

[Search These Pages] [DAI Home Page] [Comment]