TY - CONF

T1 - Iterative vs Simultaneous Fuzzy Rule Induction

AU - Galea, Michelle

AU - Shen, Qiang

N1 - M. Galea and Q. Shen. Iterative vs Simultaneous Fuzzy Rule Induction. Proceedings of the 14th International Conference on Fuzzy Systems, pages 767-772.

PY - 2005

Y1 - 2005

N2 - Iterative rule learning is a common strategy for fuzzy rule induction using stochastic population-based algorithms (SPBAs) such as Ant Colony Optimisation (ACO) and genetic
algorithms. Several SPBAs are run in succession with the result of each being a rule added to an emerging final ruleset. Each
successive rule is generally produced without taking into account the rules already in the final ruleset, and how well they may interact during fuzzy inference. This popular approach is compared with the simultaneous rule learning strategy introduced here, whereby the fuzzy rules that form the final ruleset are evolved and evaluated together. This latter strategy is found to maintain or improve classification accuracy of the evolved ruleset, and simplify the ACO algorithm used here as the rule discovery
mechanism by removing the need for one parameter, and adding robustness to value changes in another. This initial work also
suggests that the rulesets may be obtained at less computational expense than when using an iterative rule learning strategy.

AB - Iterative rule learning is a common strategy for fuzzy rule induction using stochastic population-based algorithms (SPBAs) such as Ant Colony Optimisation (ACO) and genetic
algorithms. Several SPBAs are run in succession with the result of each being a rule added to an emerging final ruleset. Each
successive rule is generally produced without taking into account the rules already in the final ruleset, and how well they may interact during fuzzy inference. This popular approach is compared with the simultaneous rule learning strategy introduced here, whereby the fuzzy rules that form the final ruleset are evolved and evaluated together. This latter strategy is found to maintain or improve classification accuracy of the evolved ruleset, and simplify the ACO algorithm used here as the rule discovery
mechanism by removing the need for one parameter, and adding robustness to value changes in another. This initial work also
suggests that the rulesets may be obtained at less computational expense than when using an iterative rule learning strategy.

U2 - 10.1109/FUZZY.2005.1452491

DO - 10.1109/FUZZY.2005.1452491

M3 - Paper

SP - 767

EP - 772

ER -