Unsupervised machine learning system selection method based on metamorphic testing

Abstract

The invention discloses a method for selecting an unsupervised machine learning system based on metamorphic testing. According to the method, 11 universal metamorphic relations suitable for most application scenes are defined on the basis of general expectation of a user on a clustering system by taking dynamic characteristics of measurement data as a purpose, so that the oracle problem in the field of unsupervised machine learning is effectively relieved, and the purpose of verifying the unsupervised machine learning system is achieved; secondly, a system characteristic-oriented sufficiency criterion based on the metamorphic relationship is defined, the metamorphic relationship is endowed with different weights by a user according to importance and severity of a failure clustering mode, and a final weighted score is calculated; according to the invention, a set of complete unsupervised machine learning system verification and evaluation system is finally formed; a user is allowed to select a part of universal metamorphic relations according to own requirements or define a new specific metamorphic relation according to domain related knowledge, and finally the user is helped to deeply understand algorithm characteristics and select an unsupervised machine learning system suitable for a specific scene.

Description

technical field [0001] The invention relates to the field of computer technology, in particular to a method for selecting an unsupervised machine learning system based on metamorphosis testing. Background technique [0002] Unsupervised machine learning algorithms are widely used in the computer field. Among them, clustering algorithms can mine the internal relationships in unlabeled data sets and establish corresponding models to meet the various needs of users. However, due to the characteristics of the algorithm, it is difficult or impossible to verify the correctness of the output results, so as to select the corresponding clustering algorithm (unsupervised machine learning system). [0003] In the existing technology, there are roughly two ways to choose an unsupervised machine learning system: [0004] The first is to select through external validation techniques: the basic idea is to compare the clustering results with external benchmarks or metrics corresponding to ...

AI Technical Summary

Problems solved by technology

[0010] The present invention proposes a method for selecting an unsupervised machine learning system based on a metamorphosis test, which is used to solve or at least partially solve the technical problem that the existing method cannot select a suitable clustering algorithm or the effect is poor

Images