Concept drift data stream ensemble classification method

The invention discloses a concept drift data stream ensemble classification method, which comprises the following steps of S1, acquiring training data and to-be-classified data, training a plurality of base classifiers through initial training data to form an ensemble model M, and integrating a novel class detection module to form an ensemble model M; s2, continuously arriving data streams are classified through a semi-supervised integration model M composed of multiple clustering-based base classifiers; s3, outlier judgment is carried out on the test examples in the data flow; and S4, classifying the test examples, calculating a confidence score to estimate the confidence of classification, and storing the confidence number in a dynamic window W. A dynamic classifier boundary is used, so that possible new instances can be better screened out, and meanwhile, through concept drift detection, the model can spend less time cost and keep better classification performance; the algorithm has high accuracy to identi