Messy genetic algorithm learns a classifier to design multiplexers

A messy genetic algorithm uses an evolutionary method to produce a classifier system. This system produces designs as Boolean functions expressing solutions to multiplexer design problems. The system learns and produces solutions for 6-, 11- and 20-line multiplexers. The designs produced by an evolutionary method perform with 94%, 85% and 63% accuracy, respectively, for these three sizes of multiplexers. The results are compared with those from a simple genetic algorithm and with those from the BOOLE system of S.W. Wilson (1986). The efficiency in obtaining solutions depends on optimal selection of initial lengths of the building blocks. The mechanism that controls the probability of splicing strings is explored also. Experiments show that it is possible to limit chromosome lengths without sacrificing efficiency