A Software Defect Prediction Method Based on Transfer Learning

Abstract

The invention discloses a software defect prediction method based on migration learning. The method utilizes the idea of ​​dimensionality reduction migration learning, comprehensively considers the probability distribution and conditional probability distribution between different project data samples, and establishes a relationship between the source project and the target project. The new feature representation minimizes the difference between the two in a new space, trains a new classifier, and then realizes migration learning. The algorithm first uses a distance measure between different distributions, the maximum mean square difference to quantify the distribution difference and the conditional distribution difference between the source data and the target data, and obtains a model by minimizing this measure. Through this model The mapped training data and test data have almost the same probability distribution and conditional probability distribution; then traditional machine learning algorithms can be used to classify the test data.

Description

technical field [0001] The invention relates to the technical field of software engineering, in particular to a software defect prediction method based on migration learning. Background technique [0002] In the past 30 years, software defect prediction has gradually become a significant research direction, dedicated to estimating how many remaining defects exist in the west facade of a software system. Software defect prediction can timely and accurately predict whether software modules contain defects in the early stage of system development, reasonably allocate test resources, and analyze defect modules in a targeted manner to improve product quality. In recent years, with the development of statistical learning and machine learning technology and its excellent prediction performance, software defect prediction methods based on statistical learning methods and machine learning methods have been gradually adopted by researchers and become the mainstream defect prediction t...

AI Technical Summary

Problems solved by technology

[0003] However, these classification techniques usually face a problem: when it is necessary to predict defects for new items, there are too few training samples, and it is difficult to train the correct model

The disadvantage of this method is that the dissimilar data discarded during the sample selection process will also cause the useful information contained in it to be discarded.

However, judging from the existing cross-engineering prediction performance, this method cannot produce better prediction results

Images