Implementation of a genetic algorithm based associative classifier system (ACS)

Abstract

The first results from the development of a genetic algorithm-based ACS are presented. The ACS is a result of mapping the inherent parallelism in classifier systems to a program which executes on a PC-based associative processor. The associative algorithms of the ACS for the coherent processor are presented. It is demonstrated that this associative implementation of the BOOLE classifier system learns as well as results published for serial implementations. It is shown that the use of an associative processor as a co-processor can decrease classifier system response time, particularly for classifier systems with a large number of rules. In fact, when the number of rules in the ACS was increased by an order of magnitude, the response time of the system increased only 25% after DOS overhead was removed.<>