Evolutionary algorithms for constructing linguistic rule-based systems for high-dimensional pattern classification problems

In this paper, we show how evolutionary algorithms can be utilized for constructing linguistic rule-based systems for high-dimensional pattern classification problems with many continuous attributes. Difficulty in handling a high-dimensional problem arises from the exponential increase of the number of linguistic rules with the dimensionality of the pattern space. For example, when we have six linguistic labels for describing each attribute in a 10-dimensional problem, the total number of linguistic rules is 6¹⁰≅6.7×10⁷. For constructing a linguistic rule-based system, we have to find a compact rule set from such a large number of linguistic rules. In this paper, we examine two approaches of evolutionary algorithms. One is a GA-based rule selection method where a small number of linguistic rules are selected from a large number of candidate rules by genetic algorithms. A subset of candidate rules is handled as an individual in this approach. The other approach is a classifier system where each linguistic rule is handled as an individual. These two approaches are compared by computer simulations on a 13-dimensional pattern classification problem