Improved XCS in Classification Problems

Abstract

The paper presents the first results of the Improved XCS in classification problems. We classify the classifiers into certain-right classifiers, certain-wrong classifiers and uncertain classifiers, and then analyze the difference between certain and uncertain classifiers. A new classifier attribute at is presented which denotes the ascending times of prediction error, the bad condition set is used to improve the robustness of the system. Several experiment results on binary and integral classification problems demonstrate the superior performance of the Improved XCS.