Abstract

In the paper the role of new expert and expert retirement conditions are analysed. In this paper we take Kolter/Maloof algorithm as a basis. The main our suggestions are: 1) variable new expert weight with expert retirement in order to overcome relative weight reduction problem of the “right” experts; 2) diversity based old expert retirement. Such option lets to achieve the same (sometimes better) classification accuracy as infinite expert system but using lower computational resources and in cases this outperforms “the oldest” and “the weakest” expert retirement methods.