Genetic Algorithm Based Path Testing: Challenges and Key Parameters

Although many studies have used Genetic Algorithms (GA) to generate test cases for white box software testing, very little attention has been paid to path testing. The paper aims to expose some of challenges posed by path testing, and to analyze what control parameters most affect GA's performance with respect to path testing. Each step in path testing is analyzed based on its complexity and automation. Experiments consist of running GA-based path testing on 12 test problems taken from the literature, using different combinations of values for important control parameters (population size, number of generations, allele range, and mutation rate). The results show that population size matters most in terms of path coverage and number of fitness evaluations, followed by allele range. Changing number of generations or mutation rate has less impact. We also make some observations about what sorts of paths are most difficult to cover. The understanding gained from these results will help to guide future research into GA-based path testing.