Automated Test Data Generation Based on a Genetic Algorithm with Maximum Code Coverage and Population Diversity

In the present paper, we investigate an approach to intelligent support of the software white-box testing process based on an evolutionary paradigm. As a part of this approach, we solve the urgent problem of automated generation of the optimal set of test data that provides maximum statement coverage of the code when it is used in the testing process. We propose the formulation of a fitness function containing two terms, and, accordingly, two versions for implementing genetic algorithms (GA). The first term of the fitness function is responsible for the complexity of the code statements executed on the path generated by the current individual test case (current set of statements). The second term formulates the maximum possible difference between the current set of statements and the set of statements covered by the remaining test cases in the population. Using only the first term does not make it possible to obtain 100 percent statement coverage by generated test cases in one population, and therefore implies repeated launch of the GA with changed weights of the code statements which requires recompiling the code under the test. By using both terms of the proposed fitness function, we obtain maximum statement coverage and population diversity in one launch of the GA. Optimal relation between the two terms of fitness function was obtained for two very different programs under testing.

Automated summary

This paper investigates an approach to intelligent support of software white-box testing process using an evolutionary paradigm, solving the problem of automated generation of optimal test data set. By formulating a fitness function with two terms and implementing genetic algorithms, it is possible to achieve maximum statement coverage and population diversity in one launch of the GA. The optimal relation between the two terms of fitness function was obtained for two different programs under testing.