Progressive integration classification method based on data with noisy tag

TL;DR: In this article, a progressive integration classification method based on data with a noisy tag is proposed, which consists of a training sample and a test sample are inputted; sample dimension sampling is carried out by using a bootstrap method to obtain B bootstrap branches; classifier training was carried out on the B-bootstrap branches by using an LDA method; an empty integrated classifier set gamma (P) is established newly and a first classifier is selected from the generated classifiers and the selected first classifiers are added into the gamma(P), classifiers meeting conditions

Abstract: The invention discloses a progressive integration classification method based on data with a noisy tag. The method comprises: a training sample and a test sample are inputted; sample dimension sampling is carried out by using a bootstrap method to obtain B bootstrap branches; classifier training is carried out on the B bootstrap branches by using an LDA method; an empty integrated classifier set gamma (P) is established newly and a first classifier is selected from the generated classifiers and the selected first classifier is added into the gamma (P); classifiers meeting conditions are selected from the rest of classifiers step by step and the selected classifiers are added into the gamma (P); selection is not stopped unless the selected number reaches a preset number G; a selected integrated classifier set and weights corresponding to all classier branches are outputted; and the test sample is classified and a final prediction result is obtained. In a data set with a noisy tag, the sample dimension and the attribute dimension are studied simultaneously, so that a good classification result is obtained.