Research On Generating Test Cases Method Based On Support Vector Machine

In the field of software testing,the design and generation of qualified test cases is a research hot spot.In the process of functional test,with the increasing of functional requirements,the scale of software increases gradually as well as the structure becomes more complex,which may cause lots of hidden bugs.Although there are many test data generator can generate plenty of test cases automatically,these test cases are lack of desired results.However,it is not only expensive,but also tedious and error prone to design the expected results for each test case manually.The background and basic concept of the software testing are introduced in this thesis,the basic principle and relevant knowledge of Support Vector Machine(SVM)are involved also.The functional testing methods are summarized and the application of machine learning in software testing are emphatically introduced.The test tool(PICT)and its combination coverage criteria are studied,which help to establish the functional model of software under test,and evaluates its performance.With the research basis of the automatic generation with input data,SVM model is set up according to the input/output relationship of the software under test,which can be used to predict the expected result of test cases.Hence,a novel method is proposed to generate test case with the expected output result in this thesis.The typical sample set comes form 2-way or 3-way combination of input parameters are selected to level docking the output result upon the relationship of input and output in the software under test.With the method of artificial,the typical sample set covers all combination of output parameters,which is used to train SVM.The model can predict the expected output from n-way input data set.Finally,the method is applied to the example of "model file validation of the activity diagram",Accuracy and Precision indicators are used to evaluate the predicted results.The experiment results show that a functional model with better performance is built with genetic algorithm,which can optimize the parameters and taking into account the combination of parameters in the input field and the output field.The method has been proven effective.