Skip to Main Content
This paper proposes a methodology to generate artificial data sets to evaluate the behavior of machine learning techniques. The methodology relies in the definition of a domain and the generation of data sets from this domain by means of different sampling processes. Then, learners are trained with the generated data sets and the created models are compared with the original domain to evaluate the quality of the learners. In the present work, a particular implementation of this methodology is provided, which is defined to test learning techniques that use a binary rule knowledge representation. As a case study, the behavior of XCS, the most influential learning classifier system, is analyzed following the methodology.