Grammatical evolution for classification into multiple classes

Abstract

In this contribution the authors deal with classification problems using an approach based on grammatical evolution. The named method is used to create short executable structures which are evolved to classify given input into multiple classes. Resulting structures are usable as computer programs for embedded devices with low computational resources. An universal formula for fitness value calculation of the evolved individual is introduced and an example of planar graphical objects classification in generated image dataset is presented. The presented approach is still applicable for general multi-class classification problems. The results of the proposed method are discussed and examined.