Statement spectrum with two dimensional eigenvalues for intelligent software fault localization

Abstract

The existing Software Fault Localization Frameworks (SFLF) based on program spectrum for estimation of statement suspiciousness have the problems that the feature type of the spectrum is single and the efficiency and precision of fault localization need to be improved. To solve these problems, a framework 2DSFLF proposed in this paper and used to evaluate the effectiveness of software fault localization techniques (SFL) in two-dimensional eigenvalues takes both dynamic and static features into account to construct the two-dimensional eigenvalues statement spectrum (2DSS). Firstly the statement dependency and test case coverage are extracted by the feature extraction of 2DSFLF. Subsequently these extracted features can be used to construct the statement spectrum and data flow spectrum which can be combined into the optimized spectrum 2DSS. Finally an estimator which takes Radial Basis Function (RBF) neural network and ridge regression as fault localization model is trained by 2DSS to predict the suspiciousness of statements to be faulty. Experiments on Siemens Suit show that 2DSFLF improves the efficiency and precision of software fault localization compared with existing techniques like BPNN, PPDG, Tarantula and so fourth.