A Multi-Resolution Learning Approach to Tracking Concept Drift and Recurrent Concepts
Mihai Lazarescu · 2005
Abstract. This paper presents a multiple-window algorithm that combines a novel evidence based forgetting method with data prediction to handle different types of concept drift and recurrent concepts. We describe the reasoning behind the algorithm and we compare the performance with the FLORA algorithm on three different problems: the STAGGER concepts problem, a recurrent concept problem and a video surveillance problem. 1