Transfer Estimation of Concept Drift in Data Stream Classification
Sun Yue-yao · Tongji yu xinxi luntan · 2011
The concept drift of data stream classification is the forefront and difficult of data mining technology,the focus of its is possible on the drift phenomenon with the hierarchical classification of data transfer sequence.Although the existing dynamic classification algorithms of concept drift have been proposed,they are not very good about the distribution of existing algorithms as examples of the lack of goals in the estimating performance of the concept drift,the number of examples seriously impacts on the estimated parameters.Based on this,a new parameter estimation method,known as the transfer estimation,is presented in this paper,using the target distribution data and similar distribution theory to improve on existing algorithms,the concept drift is correctly detected and estimated in the classification of data streams,The Improved algorithm in the data stream classification is superior to existing algorithms in the estimation of concept drift.