Iterative rule learning is a common strategy for fuzzy rule induction using stochastic population-based algorithms (SPBAs) such as Ant Colony Optimisation and genetic algorithms. Several SPBAs are run in succession with the result of each being a rule added to an emerging final ruleset. Between SPBA runs, cases in the training set that are covered by the newly evolved rule are generally removed, so as to encourage the next SPBA to find good rules describing the remaining cases. This paper compares this IRL variant with another variant that instead weights cases between iterations. The latter approach results in improved classification accuracy and an increased robustness to parameter value changes.
|Number of pages
|Published - 2005