Classification method for unbalanced data set

Abstract

The invention discloses a classification method for an unbalanced data set, and relates to class imbalance. Aiming at the problem of class imbalance in a data source, most of the methods downsample aplurality of types of samples to cause loss of a large amount of useful information, the invention provides the classification method for the unbalanced data set, which fully and reasonably uses the plurality of types of samples based on ensemble learning and logistic regression so as to further improve the classification effect. The method comprises data preparation and model training. Each weaklearner uses a plurality of types of samples which are completely different to learn, and compared with a traditional undersampling method, the weak learner can make full use of information of the plurality of types of samples. Each weak learner uses a plurality of different types of samples, and the feature set and the training parameters certainly improve the diversity of the samples, so that the integration effect is improved. The output of each weak learner is adaptively combined by using logic regression, and compared with the traditional simple average output, the output is more reasonable and anti-interference.

Description

technical field [0001] The present invention relates to category imbalance, in particular to a classification method for imbalanced data sets. Background technique [0002] Class imbalance refers to the situation that the number of training samples of different classes in the classification task is very different. In reality, most classification tasks face unbalanced data sets. At present, there are mainly two types of mitigation methods for the problem of class imbalance in binary classification datasets: the first type is to directly "under-sample" the majority class samples in the training set, that is, to remove some majority class samples so that the number of classes is close to [1-4] , and then learn; the second type is to "oversample" the minority class samples in the training set [5-6] , that is, increase the number of minority class samples to make the number of classes close, and then learn. Oversampling methods are prone to overfitting, especially when the clas...

AI Technical Summary

Problems solved by technology

[0017] The purpose of the present invention is to solve the problem of class imbalance in the data source. Most methods down-sample the majority class samples, resulting in the loss of a large amount of useful information, and provide sufficient and reasonable use of the majority class samples based on integrated learning and logistic regression, thereby A classification method for imbalanced data sets to further improve the classification effect

Images