Research On Software Defect Location Method Based On Feature Analysi

As the scale and complexity of software increase,how to ensure software quality has become a focus of researchers.Software debugging is an important means to detect software faults and ensure software quality,among which fault localization is the most time-consuming and laborious link in the software debugging process.Spectrum-Based Fault Localization（SBFL）is a class of software fault localization methods that are widely used.This class of methods analyzes the coverage information and execution result information of test cases on program entities to locate faults in the program more accurately,and assists programmers to complete code debugging.At present,SBFL has achieved good fault localization performance.However,even using the SBFL technique DStar,which has achieved the best localization effect in many empirical studies,to locate faults in the programs in the Defects4J dataset,only 8.33%of the faults can be located by checking the suspicious statement ranked first.Therefore,the localization effect of SBFL still needs to be further improved.By analyzing the statement coverage information matrix,it is found that there is a serious data redundancy phenomenon in it,which will increase the cost of fault localization and reduce the fault localization accuracy of SBFL.In addition,in the process of SBFL fault localization,only the coverage information and execution results of test cases are usually used,and the semantic and structural information implied in the code are not fully utilized,which will also affect the localization effect.Based on the existing work,this paper studies the software fault localization method based on feature analysis.First,aiming at the serious data redundancy problem in the statement coverage information matrix,a fault localization method Fault Localization based on Redundant Coverage Information Reduction（FLRR）is proposed.By using feature selection technology to reduce the statement coverage information matrix,a subset of statement coverage information matrix is obtained.The suspiciousness values of the remaining statements in the subset are calculated for fault localization,and an attempt is made to reduce the redundancy of spectrum coverage information to improve the performance of fault localization.The experimental results on Defects4J dataset show that FLRR improves at least 38.24%on E_inspect@1 indicator compared with baseline techniques such as DStar and Ochiai.Then,aiming at the problem that the semantic and structural information implied in the code are not fully utilized,a software fault localization technique Fault Localization based on combing static Features and Spectrums（FLFS）is proposed.The selected metric element indicators are modified to be suitable for measuring method-level features of programs,and static features of each method in programs are extracted for training fault prediction models.By integrating spectrum suspiciousness and prediction suspiciousness,faults are located in methods.The experimental results based on Defects4J dataset show that FLFS improves 13.22%and 7.63%on E_inspect@1 indicator compared with DStar and Ochiai respectively.