Artificial recurrence for classification of streaming data with concept shift

Abstract

The article presents a method for improving classification of streaming data influenced by concept shift. For this purpose the algorithms designed for recurring concept drift environments are adapted. To minimize classification error after concept shift, an artificial recurrence is implemented serving as a better starting point for classification. Three popular algorithms are tested on three different scenarios and their performance is compared with and without the application of an artificial recurrence. © 2011 Springer-Verlag.