Genetic Algorithm-based Software Testing Resource Allocation Problem

By the widely application of software industry in various fields of our society, people's requirement in software reliability is kept increasing. As the most important way in ensuring software reliability, software testing is getting more and more attentions. Obviously investing more testing resource is an efficient way in improving the outputs of software testing, however, in practical software developments, the limit of testing resource is a problem that the team would have to face, and there is an unavoidable conflict between cost of development and software reliability too. Therefore, that how to make a balance between software reliability and its testing cost under limited resources, becomes an important issue in software development.In the past 2 decades, there were plenty of researchers who tried to find a strategy of allocating testing resource to solve this problem through their research on unit testing process of modularized software. The researchers expected to make an optimization with some constraints of cost or software reliability, which was proved to be efficient in practical work. However, by the persistent growing of software scale, the techniques in software development were improved, the software structure were getting more complex. The requirement in software testing couldn't be satisfied by just researching on a pure unit-testing level. As a result, researchers began their explorations in the field of balancing cost and reliability of complex software model by allocating testing resource.Aiming at the issue below, our research focused on a case of hybrid structure with multi-modules, and built a testing cost model based on expected total cost and risking cost. A weighted sum associated with genetic algorithm was imported to make an optimized solution. Furthermore, we also established a multi-objective optimization strategy based on preference of decision maker handled by an interactive genetic algorithm associated with fuzzy pattern recognition. The main point of the research is illustrated in this paper as follow. 1. The discussion on the model of testing cost based on the expected testing cost and risk cost of hybrid structure software;2. The solution to the multi-objective optimization problem using genetic algorithm based on weighted sum, a research on the impacts of weighted parameter and testing redundancy upon the simulations, which supplied an efficient advice for evaluating the parameter, was built too;3. A multi-objective optimization strategy based on the decision maker's preference, which would improve the practicability of the optimization work, was established by researching on the interactive genetic algorithm and fuzzy pattern recognition.