A differential evolution based approach to generate test data for data-flow coverage

Abstract

Software testing is a vital and an effort intensive phase of the software development process. Testing efficacy relies upon optimal test data from the input domain of the problem in accordance with a test adequacy criterion. Search-based evolutionary algorithms have been widely applied for automated test data generation; Genetic Algorithm (GA) and its variants being the choice of researchers. Other highly-adaptive evolutionary algorithms such as Particle Swarm Optimization (PSO) and Differential Evolution (DE) have been shown to be more accurate and efficient in comparison to GA. This paper presents a DE-based approach to generate optimal test data in accordance to the data-flow coverage test adequacy criterion. Fitness function is designed based on the concepts of dominance relations and branch distance metric. Measures such as average number of generations and average percentage coverage are collected and analysed to evaluate the performance of the proposed approach and for comparison with Random search, GA and PSO techniques on a set of benchmark programs. The results obtained have shown that the proposed DE-based approach is competent and have better performance than random search, GA and PSO with respect to optimal test data generation in accordance to the data-flow coverage test adequacy criterion.