A cross-project software defect prediction method based on a twin network

By employing a data migration and differential weight fusion method based on Siamese networks, the problems of data distribution differences and class imbalance in cross-project software defect prediction are addressed, achieving higher prediction accuracy and stability, and enhancing the model's ability to identify defect samples.

CN122364092APending Publication Date: 2026-07-10HARBIN ENG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610590216.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-04-30
Publication Date
2026-07-10

AI Technical Summary

Technical Problem

In existing cross-project software defect prediction technologies, there are significant differences in data distribution between the source project and the target project, and a severe imbalance between defect-type samples and non-defect-type samples. This makes it difficult for the model to effectively transfer defect-related features and learn insufficiently from a few defect samples, thus failing to guarantee prediction accuracy and stability.

Method used

A twin network-based approach is adopted, which addresses the issues of data distribution differences and class imbalance across projects through data migration, difference weight fusion, and twin network feature encoding, including data normalization, training of a GAN-based software defect data migration model, difference weight calculation, and training of a Deep SVDD defect discriminator.

Benefits of technology

It improves the accuracy and stability of cross-project software defect prediction by preserving the defect-related features of the source project, enhancing data adaptability, reducing noise risk, mitigating class imbalance, and improving the model's ability to identify defect samples.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122364092A_ABST
    Figure CN122364092A_ABST
Patent Text Reader

Abstract

A cross-project software defect prediction method based on Siamese networks is proposed. This method addresses the problems in cross-project software defect prediction, such as significant differences in data distribution between the source and target projects, and imbalance between defect and non-defect sample categories, which hinder effective transfer of defect features and result in insufficient prediction accuracy and stability. The proposed solution involves: uniformly normalizing both the source and target datasets; training the software defect data transfer based on generative adversarial networks; calculating the difference weights between the source and target datasets; fusing the transferred data with the difference weights to generate a weighted augmented dataset; pre-training the dataset and combining it with Deep SVDD to construct a defect discriminator; and finally, obtaining a cross-project software defect prediction model. This model is applicable to cross-project defect prediction and high-risk module identification during software development, testing, and maintenance.
Need to check novelty before this filing date? Find Prior Art