Generating test data for black-box testing using genetic algorithms

Abstract

Software testing is the most important and most time-and cost consuming part in the software development process. According to [10] about 50% of the resources in a software project are spent for testing. Often, research activities, with the goal to automatise this process, cover only the generation of test cases and their execution. The important topic of generating meaningful test data is often left out. Approaches based on the analysis of source code exist, but are inapplicable to perform black-box testing or test message-based services with more complex data types. Thus, the test data is usually gathered either manually by a test developer or randomly. The manual way requires a lot of experience on developers' side and the random approach produces a large amount of redundant test data covering identical use cases. This paper proposes a new approach to automatically generated test data for black-box testing exploiting genetic algorithms (GA). To enhance the quality of the test data micro genetic algorithms are used as a filter. With this approach a test developer can produce a limited number of qualitative test data in a controlled way, even if the source code is not available. This paper will examine the different components of a GA and what requirements they must fulfill to be suitable for a test data generation.