Study On Neural Network And Its Application In Software Fault Localization

Software debugging is a very important part of the software development process, which is also a time-consuming and laborious task. Any improvement in fault localization in software debugging will greatly reduce the cost of software debugging and maintenance, and of course has very important significance. In recent years, automated software fault localization technology has attracted many scholars’ attention, various approaches have been proposed and considerable development has been made in terms of techniques and tools. These techniques utilize coverage information from running programs against test cases, and locate the faults in the program by comparing differences in program runs. Being a machine learning method, BP neural network is a commonly accepted, widely used in a lot of fields and has successfully solved some problems in reality.On the basis of surveying several typical fault localization techniques, we improve the standard BP algorithm and apply this variance to software fault localization, and thus propose an improved technique of neural network-based fault localization. Through processing the program running traces, raw data are converted as the training samples of BP neural network. Data are orderly input into BP neural network in training, the weights of the network are adjusted repeatedly until the network error rate is within the preset range, and finally test matrix is used to compute the suspiciousness of each statement, and faults are assumed to be located at the statements with higher suspicious value.Using MATLAB R2009b for the purpose of experiment, we build the neural network by calling the neural network function in the neural network toolbox of MATLAB R2009b, and write programs to calculate suspicious value of each statement. In order to verify the validity of the algorithm, we adopt the Siemens program as the experimental benchmark, and some of the existing technologies are elected for comparison. The experiments show that the improved-BP neural network-based fault localization technology has certain validity.