Cooperative Evolution of Rules for Classification

Abstract

A new learning technique based on cooperative coevolution is proposed for tackling classification problems. For each possible outcome of the classification task, a population of if-then rules, all having that certain class as the conclusion part, is evolved. Cooperation between rules appears in the evaluation stage, when complete sets of rules are formed with the purpose of measuring their classification accuracy on the training data. In the end of the evolution process, a complete set of rules is extracted by selecting a rule from each of the final populations. It is then applied to the test data. Some interesting results were obtained from experiments conducted on Fisher's iris benchmark problem