CLASSIFIER GENERATION METHOD USING COMBINATION OF MINI-CLASSIFIERS WITH REGULARIZATION AND USES THEREOF

摘要

A kind of method,Generating for classifier includes a step, and the classification of the data for obtaining multiple samples, each sample of data includes multiple physical measurement characteristic values and class label. Each miniature classifier, which generates, uses two sets of features from sample. The performance of miniature classifier, it is retaine that test, which meets performance threshold with those,. Training and tag along sort sample of the Main classification device by a set of small sortation devices of conduction regularization assemblage retained/filtered, for example, carrying out the miniature classifier of logistical training of this selection by the miniature classifier of filtering (regularization of dropping out) of random selection fraction. This group of sample is randomly divided into test set and training set. It generates and repeatedly generates Main classification device with the miniature classifier, filtering and test and training set is entered for the separating sample group of different realizations, to generate multiple Main classification devices. Final classification device defines the combination Main classification device from one or more than one.