Combining non revisiting genetic algorithm and neural network to generate test cases for white box testing

Abstract

Software testing consumes 50% of the total expenditure done on the overall software. Cost of the testing phase is generally high due to heavy manual intervention. Many steps have been made in literature to reduce the cost of the phase by replacing existing manual work with automatic process. For example, manual test case generation is replaced by automatic test generation. A number of search methods exist in literature to automate test generation process in which Genetic Algorithm based approaches are very popular. A number of papers have been proposed in literature to show the use of GA to generate suitable test data for given software for desired adequacy criteria. Although GA is very effective in searching good test cases yet it has some inherent weaknesses. For example application of Simple Genetic algorithm may generate redundant test data in upcoming generation. This requires additional effort and expenditure without adding any value to the desired objective. This problem can be eliminated if non-revisiting GA is used in place of simple GA. GA also use fitness of various test cases to judge when to finished searching. At present this fitness evaluation is done by a human being hence it is a costly affair. This work can be automated by replacing a human being with a neural network. This paper proposes a combination of neural network and genetic algorithm to overcome the existing problems of automatic test data generation for white box testing. © 2011 Springer-Verlag Berlin Heidelberg.