A Multi-objective Competitive Co-evolutionary Approach To Test Case Prioritization

As a significant part of software life cycle, regression testing is an effective measure to guarantee the quality of software products. In many software development projects, the expense of regression testing accounts for more than 50% of the entire budget. In order to reduce the cost as much as possible, the researchers from different countries have conducted extensive studies, in which the test case prioritization has become a research focus in recent years.However, there are some problems that have been exposed in the traditional prioritization technique:the first shortage is that the traditional prioritization techniques use the single test criterion to guide the entire ordering process, unable to meet the constraints of multiple objective factors; the second shortage is the traditional prioritization techniques have slow convergence and easy to fall into local optimum.To cover the shortage of single object genetic algorithm in test case prioritization, a new competitive co-evolutionary approach was adopted to address these problems. In this new approach, multiple metrics of fitness had been used, including the absolute fitness that evaluated the survival ability of an individual and the relative fitness that estimated the number of defeated opponents of each individual. The outstanding individuals who defeated more opponents can join the elite set for further evolution. By eliminating "old" individuals, the individual’s survival time is used to avoid the local optimum. To improve the efficiency of error detection, our study introduced the average percentage of mutation kill rate as a new optimization criterion. Comparing to the classical search algorithm, the Co-evolutionary Algorithm improved the search efficiency and local search ability, and can be applied for different scales of programs to obtain a higher average percentage fault detection rate and a uniform distribution of the Pareto boundary. The experiments verified the validity and the practical value of the new approach.