Classification method for data stream based on concept drift detection technique

This paper proposes a new classification method for data stream based on the combination concept drift detection and classification model. The proposed method includes a pooling mechanism, which stores classifiers corresponding to different concepts to ensure that the classification model will not do re-training when those concepts which appeared previously are present again, so as to directly sort out the appropriate classifiers from the pool to classification. At last, it overviews different concepts and finds out the transition relationships among them and visualizes them.