一种新的基于机器学习的缺陷定位方法

A Novel Fault Localization Method Based on Machine Learning

缺陷定位是软件调试过程中的重要部分,在这个过程中往往需要花费大量的人力和时间。因此,如何自动且精确地对被测软件进行缺陷定位是目前许多学者关注的问题,其中一些学者将程序频谱与机器学习技术相结合进行缺陷定位,并取得了一些成果。然而,由于在程序频谱的构建中往往只考虑到程序中语句是否被覆盖,并未考虑其被覆盖的次数,这将丢失掉一些程序运行时的信息,从而对定位模型的定位效果造成一定的影响。本文通过对程序频谱的构造技术进行改良并将其与机器学习技术结合,提出了一种新的缺陷定位方法IML,然后将IML与已有的基于机器学习的缺陷定位方法在三个西门子套件测试程序上进行了实验对比。实验结果表明,IML较已有的基于机器学习的缺陷定位方法能取得更好的缺陷定位效果。

Fault localization is an important part of software debugging process,which often requires a lot of manpower and time.Therefore,how to locate fault statement automatically and accurately is a problem that many scholars pay attention to solve.Among them,some scholars combine program spectrum with machine learning technology to locate fault statement,and they have achieved some results.However,when construct program spectrum,it often only considers whether the statements in the program are covered,but does not consider the number of times they are covered.it will lose some information,it will be affecting the performance of the fault localization model.Therefore,in this paper,we will improve the construction technology and combine it with machine learning technology to propose a novel fault localization method IML.Then we compare IML with the existing fault localization method based on machine learning in three Siemens test suit.The experimental results show that IML is better than the existing fault localization method based on machine learning in fault localization.