Comparison of Strategies Based on Evolutionary Computation for the Design of Similarity Functions

Abstract

One of the main keys in case-based reasoning system is the retrievalphase, where the most similar cases are retrieved by means of a similarity function.According to the problem, the similarity function must be selected and adapteddepending on the characteristics and properties of the problem’s domain. The goalof this article is to present a platform called BRAIN, which incorporates strategiesbased on different evolutionary approaches to design similarity functions ad hocfor a domain to be used in a case-based reasoning system. The strategies are basedon Genetic Programming and Grammar Evolution approaches. Both are applied todifferent data sets to study the influence of their characteristic in the accuracy rateand in the execution time.