Method and device for predicting abnormal sample

Abstract

The embodiment of the specification provides a method and a device for predicting an abnormal sample. The method comprises the following steps that: firstly, obtaining a sample to be predicted, then,adopting a plurality of dimensionality reduction methods to independently carry out dimensionality reduction processing on the sample to be predicted to obtain a plurality of processed samples; then,independently inputting the plurality of processing samples into a plurality of processing models to obtain the score of each processing sample, wherein the ith processing model Mi scores a corresponding processing sample on the basis of a hypersphere determined by a SVDD (Support Vector Data Description) way in a corresponding dimensionality space in advance to score the corresponding processingsample; then, according to the score of each processing sample, determining the comprehensive score of the sample to be predicted; and finally, according to the comprehensive score, determining whether the sample to be predicted is an abnormal sample or not. Therefore, whether an unknown sample is an abnormal sample or not can be more effectively predicted.

Description

Technical field [0001] One or more embodiments of this specification relate to the field of sample classification using a computer, and in particular to a method and device for predicting abnormal samples. Background technique [0002] With the development of computer and Internet technology, a large number of data and samples have been produced. In many scenarios, it is necessary to classify these data and samples, such as distinguishing whether it is a normal sample or an abnormal sample. For example, in payment and transaction services, it is often necessary to distinguish between normal transaction samples and abnormal transaction samples (for example, cash out, financial fraud transactions, etc.), so as to better prevent payment risks. In the field of security access, it is often necessary to distinguish between normal access data and abnormal access data. The abnormal access data often comes from some users trying to invade or obtain illegal data through illegal access. S...

AI Technical Summary

Problems solved by technology

[0004] However, in many cases, abnormal samples are often difficult to collect and calibrate

On the one hand, the number of abnormal samples is usually small. On the other hand, abnormal samples are often very hidden and difficult to be found. For example, abnormally accessed data is usually difficult to detect.

Therefore, the number of abnormal samples that can be obtained and identified is very small, which makes supervised learning difficult

Images