Empirical Evaluation of Metaheuristic Approaches for Symbolic Execution based Automated Test Generation

This paper empirically evaluates four meta-heuristic search techniques namely particle swarm optimization, artificial bee colony algorithm, Genetic Algorithm and Big Bang Big Crunch Algorithm for automatic test data generation for procedure oriented programs using structural symbolic testing method. Test data is generated for each feasible path of the programs. Experiments on ten benchmark programs of varying sizes and complexities are conducted and the subsequent performance results are presented. All the four algorithms have been evaluated on average test cases per path and average percentage coverage per path. It has been observed that the particle swarm optimization based algorithm outperforms the other three algorithms. The result also concludes that predicates solving difficulty (such as constraints having equality operator‘&&’ as join operator) has a direct relationship with testing efforts rather than program complexity measures such as cyclomatic complexity, number of decision nodes etc. Keyword: Software testing, Symbolic execution, Genetic Algorithm, Swarm Intelligence, Particle Swarm optimization, Artificial Bee Colony algorithm, Big Bang Big Crunch Algorithm.