A Parallel Evolutionary Algorithm for Discovery of Decision Rules

In the paper a new parallel method for learning decision rules is proposed. The method uses evolutionary algorithm to discover decision rules from datasets. We describe a parallelization of the algorithm based on master-slave model. In our approach the dataset is distributed among slave processors of a parallel system. The slave procesors compute fitness function of chromosomes in parallel. The remainder of evolutionary algorithm i.e. selection and genetic search operators is executed by the master processor. Our method was implemented on a cluster of SMP machines connected by Fast Ethernet. The experimental results show, that for large datasets it is possible to obtain a significant speedup.