Combination of information entropy and ensemble classification for detecting concept drift in data stream

Abstract

Concept Drift1 is considered a challenging problem that appears in data streaming. The classifier's error rate and the ensemble are used in most of the previous works to manage classification accuracy as a criterion for judging whether concept drift is happening or not. Information entropy is an effective way of measuring uncertainty and it is suitable to detect concept drifts in a reliable, fast, and computationally efficient way. This paper presents a model for dealing with two main kinds of concept drifts: Sudden and Gradual drifts in labelled data. The learning procedures of the model are processed in blocks of the same size. The proposed model called Entropy Based Ensemble (EBE) is based on incorporating the Entropy as a drift detector into the evolving ensemble. Performance of EBE was evaluated experimentally on synthetic dataset with different types of concept drifts. The primary results showed improving the accuracy of EBE in presence of concept drift.