An investigation into adaptive search techniques for the automatic generation of software test data
The focus of this thesis is on the use of adaptive search techniques for the automatic generation of software test data. Three adaptive search techniques are used, these are genetic algorithms (GAs), Simulated Amiealing and Tabu search. In addition to these, hybrid search methods have been developed and applied to the problem of test data generation. The adaptive search techniques are compared to random generation to ascertain the effectiveness of adaptive search. The results indicate that GAs and Simulated Annealing outperform random generation in all test programs. Tabu search outperformed random generation in most tests, but it lost its effectiveness as the amount of input data increased. The hybrid techniques have given mixed results. The two best methods, GAs and Simulated Annealing are then compared to random generation on a program written to optimise capital budgeting, both perform better than random generation and Simulated Annealing requires less test data than GAs. Further research highlights a need for research into the control parameters of all the adaptive search methods and attaining test data which covers border conditions.