关键词: 数据流, 集合分类器, 偏倚抽样, 偏差方差分解

Abstract:

High speed data stream brings the phenomenon that the data rate is higher relative to the ensemble classifiers computational power, so the ensemble classifiers cant train all data which reached recently to update themselves. An ensemble classifiers is proposed based on biased sample. Expectation error is analyzed through biased variance decomposition method, and the data is also biased sampled by computing all datas expectation error contribution degree which is waited for being sampled. This method can reduce time to train and update ensemble classifiers and will be contrasted with random sample ensemble classifiers. It indicates that this method has more prediction accuracy on condition the same proportion of sample.

Key words: data stream, ensemble classifiers, biased sample, bias variance decomposition