Defect high-risk module identification method based on software network

Abstract

The invention provides a defect high-risk module identification method based on a software network, and belongs to the field of software complex networks. The method comprises the steps that 1, constructing an adaptive classifier, whrein the adaptive classifier comprises a plurality of classifiers; step 2, performing adaptive feature selection; step 3, performing adaptive threshold optimization; step 4, optimizing the internal parameters of the adaptive classifier; and 5, selecting a self-adaptive optimal prediction model, and performing defect high-risk module identification on the to-be-tested software network by using the optimal prediction model. For any type of defect data set, contents in five aspects of construction of the self-adaptive classifier, self-adaptive feature selection, self-adaptive threshold optimization, self-adaptive classifier internal parameter tuning, selection of a self-adaptive optimal prediction model and the like can be completed according to the characteristics of a data set, the best defect prediction result is obtained, and a high-risk software module is identified.

Description

technical field [0001] The invention is applied to the field of complex software networks, and is a method for identifying high-risk modules with defects based on software networks. Background technique [0002] With the rapid development of the Internet, software has more and more functions to help people carry out various production activities. Whether the software is safe and reliable has received more attention, and how to identify the defective high-risk modules in the software as early as possible has become a hot research field. Accurate identification of modules with a high risk of defects can improve software quality and reduce development costs. [0003] A large number of studies at home and abroad have shown that 80% of software defects exist in 20% of software codes. However, in actual software testing, the principle of uniform coverage is often adopted, requiring 100% coverage of requirements and statements, which wastes a lot of testing resources in essence. ...

AI Technical Summary

Problems solved by technology

However, in actual software testing, the principle of uniform coverage is often adopted, requiring 100% coverage of requirements and statements, which actually wastes a lot of testing resources

In some software third-party tests, the hit rate of test cases is often less than 1%, or even lower

[0008] Relevant classifiers for machine learning have become increasingly mature, but not all classifier algorithms are suitable for certain types of data sets, such as linear regression algorithms. Such classifiers are less effective for those data sets that satisfy linear separability. Good, but poor performance on linearly inseparable datasets

The reason is that the defect prediction model established by it lacks adaptability to the data

Images