Unbalanced data classification method based on active learning

Abstract

The invention discloses an unbalanced data classification method based on active learning, and the method comprises the steps: randomly sampling and selecting a sample from original label-free data for marking, and taking the sample as initial training data; performing cost-sensitive learning training on the initial training data by adopting a universal machine learning model; predicting all samples which are not labeled in the original training data samples by utilizing a trained binary supervised classification model, and selecting N samples which are most uncertain according to uncertainty;respectively calculating the sum of Euclidean distances between the N samples and the central point of the trained data set, and selecting M samples from the N samples according to the descending order of the distances; marking the selected M samples, and adding the marked M samples into a training data set; performing cost-sensitive learning training on the initial training data set by using a universal machine learning model; and continuously repeating the process, iteratively circulating until the average uncertainty of the selected M samples is smaller than a set uncertainty threshold, and stopping training. According to the method, on the basis of keeping the performance of the unbalanced data classifier, the sample size of labeling can be effectively reduced, so that the labeling time and the labor cost are saved.

Description

technical field [0001] The invention relates to the field of machine learning, in particular to an unbalanced data classification method based on active learning. Background technique [0002] Unbalanced data widely exists in practical applications, such as fraudulent transactions and non-fraudulent transactions in the field of credit card transactions. Due to the unbalanced distribution of such data categories, traditional machine learning models cannot be directly used in the classification of such data. The methods to solve imbalanced data classification mainly include resampling method and cost-sensitive learning method. Among them, the resampling method is subdivided into methods such as undersampling, oversampling, and smote. [0003] Traditional machine learning training methods are mostly supervised learning methods, that is, given all samples of data in related fields and the labels of the samples, they are trained through appropriate machine learning models, and ...

AI Technical Summary

Problems solved by technology

This learning method needs to label all samples, which will lead to huge time and labor costs

Images