Evolutionary adapted ensemble for reoccurring context

Abstract

Reoccurring Context is a phenomenon being subject of interest in machine learning theory dealing with Concept Drift. Periodic reappearance of contexts naturally encourage designing classifier systems which utilizes their expertize on contexts collected in the past. The paper presents study on EAERC algorithm that gather its knowledge on appearing contexts in form of elementary classifiers which can potentially contribute in ensemble classifier system if necessary while keeping ensemble size strictly limited to ensure short response time. While unseen context appears EAERC automatically adds new classifier to the pool. © 2012 Springer-Verlag.