Adversarial Learning Based Software Defect Prediction For Long And Short Memory Networks

From the birth of software,software defects are born with it.Software defect prediction has always been an important research topic.The existence of software defects often brings unpredictable troubles and losses to our production and life,so in the current software development process,the software has to go through a long time of testing and troubleshooting before being put into use.Even in the use phase,it should be constantly tested and updated.The existing software defect prediction techniques mainly focus on the combination of abstract syntax tree(AST)and deep learning to predict software defects.In this paper,a defect prediction model based on adversarial learning(ADLSTM)is proposed for the first time in software defect prediction.The model first extracts Token vectors from the abstract syntax tree(AST)of the program,and then encodes them into digital vectors through mapping and word embedding.The numerical vector is then fed into the adverse-learning long-short-term memory network(ADLSTM)to learn the semantic and structural features of the program.In this paper,7 open source projects commonly used in the field of software defect prediction are used as data sets to carry out experiments.The experimental data show that,compared with the existing machine learning methods(DBN method,CNN method,LSTM method),in terms of the evaluation index F1(F1 SORCE),The adversity-learning based long-short-memory network(ADLSTM)method proposed in this paper not only performs better in defect prediction within projects,but also has the same advantage in defect prediction across projects.In this paper,the trained algorithm model is encapsulated into a micro-service application using container technology and deployed on K8 S platform.This application integrates the code defect prediction system based on adversarial learning long short memory network(ADLSTM).Users can upload code that has yet to be verified,and the system will detect if the uploaded code contains defects.As a result,users can predict code defects in their code at any time.