Neural Network Based Test Case Generation for Data-Flow Oriented Testing

Data-flow oriented testing plays an important role in software quality assurance. Many researches applied genetic algorithm to automatically generating test cases. However, each test case needs the run of program so as to compute its fitness value in most researches, which costs a lot. This paper proposes a neural network based approach for all-uses criterion oriented test case generation. The DU-pairs that need to be tested are calculated firstly. Then BP neural network is trained to simulate the fitness function. Finally, genetic algorithm is used to generate test cases where fitness value of each test case is evaluated with the trained neural network.