A Hybrid Particle Swarm Optimization and Differential Evolution Based Test Data Generation Algorithm for Data-Flow Coverage Using Neighbourhood Search Strategy

摘要

Meta-heuristic search techniques, mainly Genetic Algorithm (GA), have been widely applied for automated test data generation according to a structural test adequacy criterion. However, it remains a challenging task for more robust adequacy criterion such as data-flow coverage of a program. Now, focus is on the use of other highly-adaptive meta-heuristic search techniques such as Particle Swarm Optimization (PSO) and Differential Evolution (DE). In this paper, a hybrid (adaptive PSO and DE) algorithm is proposed to generate test data for data-flow dependencies of a program with a neighbourhood search strategy to improve the search capability of the hybrid algorithm. The fitness function is based on the concepts of dominance relations and branch distance. The measures considered are mean number of generations and mean percentage coverage. The performance of the hybrid algorithm is compared with that of DE, PSO, GA, and random search. Over several experiments on a set of benchmark programs, it is shown that the hybrid algorithm performed significantly better than DE, PSO, GA and random search in data-flow test data generation with respect to the measures collected. Meta-heuristic search techniques, mainly Genetic Algorithm (GA), have been widely applied for automated test data generation according to a structural test adequacy criterion. However, it remains a challenging task for more robust adequacy criterion such as data-flow coverage of a program. Now, focus is on the use of other highly-adaptive meta-heuristic search techniques such as Particle Swarm Optimization (PSO) and Differential Evolution (DE). In this paper, a hybrid (adaptive PSO and DE) algorithm is proposed to generate test data for data-flow dependencies of a program with a neighbourhood search strategy to improve the search capability of the hybrid algorithm. The fitness function is based on the concepts of dominance relations and branch distance. The measures considered are mean number of generations and mean percentage coverage. The performance of the hybrid algorithm is compared with that of DE, PSO, GA, and random search. Over several experiments on a set of benchmark programs, it is shown that the hybrid algorithm performed significantly better than DE, PSO, GA and random search in data-flow test data generation with respect to the measures collected.