Test Case Generation Model Based On Bayesian Network And Genetic Algorithms

The expansion of software to the test work has brought new problems that manualtesting is too slow, and too inefficient, so the automated testing is becoming important.How to generate test cases plays an important role in automated testing. Commonlyencountered problem is test case can not cover all path, even a small number of pathssimply do not have test cases implementation. This test case generation model based onBayesian networks (BN) and Genetic algorithms (GA) is proposed in this thesis. Andthe model has solved the main problem is that the automated test generate collection ofthese test cases which must meet the criteria covered by the full path.Firstly, the classic method of software testing including unit testing, black boxtesting, white box testing, static testing and dynamic testing is introduced in detail. Thecommon test case generation methods are introduced simply in this thesis. Then, somebasic theories and operations of GA are introduced. Some advantages and disadvantagesof GA applied in practical application are analyzed. Secondly, the basic concepts of theBayesian network, parameter learning algorithm and structure learning algorithm areintroduced. Test case generation model based on Bayesian networks and Geneticalgorithms is proposed, Bayesian network is an information representation frameworkcombining causal knowledge with probability knowledge. Topologies between networknodes are represented by qualitative information of BN and joint probabilitydistributions between network nodes are represented by quantitative information of BN.Therefore, the relationship among the parameters of the test program can be representedby quantitative information. The full path coverage criteria are met by adjusting thequantitative information. In the model proposed in the thesis, the Bayesian network hasbeen modified continually by GA as heuristic search algorithm until it achievesoptimization. The optimal Bayesian network model is used to generate test cases set.Three simulation experiments are processed in the end. The first experimentrevealed the defects of the full path coverage test cases. The second experiment is usedto generate test cases by the algorithm proposed in the thesis, and the results showed itis a feasible and effective method. The third experiment is used to verify the feasibilityof the model based on BN and GA in the complex tested program.