Cross-project Software Defect Prediction Based On Machine Learning

With the development of society,software is playing a more and more important role,and software quality has been paid more attention.In order to find the defects in the software during the software development stage,researchers try to use machine learning to predict the possible defects in the software.However,there may be some problems in the application of this technology,such as cold start and difference in data distribution.In order to solve the problems existing in this technology,the researchers proposed a cross-project software defect prediction method based on machine learning.Cross-project defect prediction uses the historical data of one or more source projects to establish the defect prediction model,and then carries on the defect prediction to the target project.At present,there are still some problems in the field of cross-project software defect prediction.This paper focuses on the negative transfer and the lack of semantic features in the current cross-project software defect prediction.The specific research contents of this paper are as follows.(1)Aiming at the negative transfer problem existing in the field of cross-project defect prediction,this paper proposes a multi-source integration algorithm.The algorithm uses a two-stage data filtering algorithm to select the source project data set and related data that are similar to the target project.With the updated weight of the source project data based on the improved TrAdaBoost algorithm,the negative transfer between the source project and the target project is reduced and the performance of the cross-project defect prediction model is improved.(2)Aiming at the problem of the lack of transferable semantics in the field of cross-item defect prediction,this paper proposes a convolution transformation neural network model.The model uses the word embedding model to transform the syntax tree nodes of the source code into a vector representation,and obtains the feature information through the convolution transformation neural network.In the output layer of the neural network,both the classification loss and the data distribution difference loss are minimized.Finally,the model uses the extracted feature information to construct a cross-project defect prediction model.The experimental results show that the two defect prediction models proposed in this paper have better overall performance compared with the classical cross-project defect prediction model.