Robustness and Evolution in an Adaptive System Application on Classification Task

In this paper, we proposed an approach to a single-step Classifier System, in which the useful population is built by progressively specializing classifiers. It has been applied to a classification task in a medical domain. To permit the system to explore alternatives without making decisions earlier in learning stages, all the classifiers that might be selected are triggered and receive the resulting reward corresponding to their action. The payoff function involves the classifier’s performance, its specificity and the system’s performance (its robustness). Genetic operators are activated with a probability which depends on the system’s robustness. During the test stages, no further learning takes place and the system’s performance is measured by the percentage of correct classification made on the second set of examples. When the measure of performance is the highest, the population is stabilized and contains the correct classifiers (the payoff function and genetic operators have no more effect on classifiers). This approach achieves convergency more quickly and makes it possible to have a final accurate population without over-specializing.