An Experience-Based Model For Test Execution Effort Estimation

As the role of software products becomes more and more important in the whole society, jointly with the competition on producing and delivering software products and services increases, it is very necessary to produce low cost high quality software in a short period. Software testing is one of the most important ways in the quality assurance. Making the reasonable and accurate estimation on software test execution effort is the crucial precondition of designing software plan and finishing testing successfully. A good software test execution effort estimation model can help test managers plan their schedules and resources more properly and much easier, and then improve the quality of their software products. But actually, the research on software test execution effort estimation is very short up to now, and also the existing models have some severe disadvantages, e.g. time consuming, strong subjectivity and so on.Based on this, first, this paper fully introduces the existing software estimation methods, and also discusses the existing software test execution effort estimation models in detail. Then this paper proposes an experience-based model for software test execution estimation effort. In this model, we characterize a test suite as a 3-dimensions vector called test suite execution vector which includes test case number, test execution complexity and its tester rank. Based on the definition and measurements on test suite execution vector, we set up an experience database. According to Regression Analysis concept, we use machine learning algorithm to train the experience data sets to find the relationship between test suite execution vector and test execution effort, and then estimate efforts for a given test suite execution vector. We evaluate this model through an empirical study in a financial software company, and get a good estimation result. Then we use this model in a real QA cycle for a project in this company, which helps test managers finish software test management work much better, such as designing test plan, allocating test resources and so on.