Selecting the Optimal Rule Set Using a Bacterial Evolutionary Algorithm

In many regression learning algorithms for fuzzy rule bases it is not possible to define the error measure to be optimized freely. A possible alternative is the usage of global optimization algorithms like genetic programming approaches. These approaches, however, are very slow because of the high complexity of the search space. In this paper we present a novel approach where we first create a large set of (possibly) redundant rules using inductive rule learning and where we use a bacterial evolutionary algorithm to identify the best subset of rules in a subsequent step. The evolutionary algorithm tries to find an optimal rule set with respect to a freely definable goal function.