A Software Defect Prediction Method Based on Cooperative Migration

Abstract

A software defect prediction method based on collaborative migration, including the following steps: 1) Expand the original source project data set into four new source project data of the same scale by combining four different standardization methods and the TCA transfer learning method at the same time 2) Construct a collaborative classifier for the target project using the software defect prediction algorithm based on collaborative migration; 3) Perform defect prediction on new samples to be predicted in the target project. The present invention selects four different standardization methods and combines them with the TCA transfer learning method to expand the source item data set, enriches the information expression of the source item data, generates a sub-classifier for each source item, and classifies the sub-classes according to the PSO algorithm The adaptive weight distribution is carried out by the classifier, so as to build a collaborative classifier to predict the defects of the samples to be tested in the target project.

Description

technical field [0001] The invention belongs to the field of software defect prediction algorithms, in particular to a software defect prediction method based on cooperative migration. Background technique [0002] Software defect prediction can be divided into intra-project defect prediction and cross-project defect prediction. In-project defect prediction requires a large number of samples known to be defective in the project, such as files, classes, and functions, etc., which are used as training sets, combined with machine learning methods to generate classifiers and then predict target samples. The cross-project defect prediction can predict the defect of the target project based on the samples of other related projects. In the actual development process, due to the fact that the target project is too new or the cost of obtaining labels is too high, there are too few training samples in the target project, and cross-project defect prediction is often required. In most...

AI Technical Summary

Problems solved by technology

However, none of the above algorithms have been used in the prediction of software defects.

[0005] In general, the current software defect prediction algorithms have the following problems: In software defect prediction, transfer learning is very important for cross-project defect prediction, how to make the transfer learning algorithm make full use of the useful information of the source project, thereby improving the target The defect prediction performance of the project; different source projects will have different prediction effects on the target project. Under the premise that it is impossible to determine which source project has the best prediction performance, how to consider it at the same time compared with one-to-one cross-project defect prediction All other related source items to improve predictive performance

Images