Research On Fault Localization Method Of Constructing Program Entity Suspicious Based On Ridge Regression

Software debugging is the most time-consuming part of the software development cycle,the purpose of software debugging is to repair the software fault or error,so the study of automatic fault location technology or auxiliary fault location technology is very necessary.The technique based on program spectrum is the most efficient of existing fault location technology in practical.However,there are obvious defects in this technique:(1)for all programs,the suspicion of program entities is calculated using a fixed skeptical formula,but for specific procedures,it is affected by various objective factors such as test cases and development language,the error may be different from other programs.(2)faulty statements tend to affect the following statements during the run of the program,but the technology does not take into account the association between the program entities,so it can not correctly assess the suspicion of certain statements,which leads to poor locating.In order to solve the above problems,this thesis proposes a learning model based on the relationship between program entities and test cases running results by using kernel ridge regression.Ridge regression is a machine learning method that can solve the complex collinearity relationship among features.therefore,it is possible to construct a unique procedure for calculating the entity's suspicious degree for each program.without loss of general this article treats a statement as a program entity,treats the coverage times of statement as a feature,treats the results of each test case as a target value,uses the program statement coverage of each test case and target value to build training matrix,finally,by using the kernel ridge regression to obtain the fitting curve of the sample training and the virtual vector to assess each program entity contains the possibility of failure.Since the result of the ridge regression method is two-valued and can not be fully adapted to the requirement of software fault location,this thesis presents a smoothing process for the test case to optimize the construction of the fitting curve.At the end of this thesis,a detailed experimental framework is designed and an automatic fault location procedure is established.According to the evaluation standard of the existing fault localization method,the Siemens test suite is used to validate the method of constructing the entity suspicion based on the ridge regression to evaluate the likelihood of the program entity.