Please use this identifier to cite or link to this item:
|Title:||Investigating search-based test generation for web applications|
Web site development
|Abstract:||Search Based Software Testing (SBST) re-frames the problem of test case generation as a search optimisation problem. The optimisation is enabled by a metaheuristic search technique, such as Hill Climbing. This is to say that, given a particular input domain (e.g. an integer parameter in a method), the task is to search through that domain looking for members which, when used as inputs in a test case will improve the quality of the test suite. The quality is defined by a fitness function which analyses test-suite adequacy depending on some criteria. SBST has been widely used as a means of designing automated tests in stand-alone applications, however little work has been done in applying SBST to design automated tests for web applications. Automating test design in web applications bears a number of challenges. This is especially true in web languages which involve dynamic typing, such as PHP. Additionally, it is also difficult to determine input data types during test execution since the inputs, in most cases are not explicitly specified. This study carried out an exploratory investigation on Search Based Test generation within web applications. This was achieved through an iterative investigation which was aimed at identifying challenges to leveraging SBST and proposing potential solutions to these challenges. We investigated each iteration by comparing three search algorithms; Random Search, Simulated Annealing and Tabu Search. We then evaluated each iteration on two e-commerce open source applications by analysing the coverage obtained from each application. In general, the search based approach which was implemented with Tabu Search surpassed the non-search based approach implementing Random Search in all of the iterations. Furthermore Tabu Search also surpassed Simulated Annealing and when Tabu Search was restarted the algorithm continued to provide an increase in coverage. Simulated Annealing displayed mixed results when compared to Random Search. However we identify this due to the annealing function, that was implemented in a general format for all applications. This suggests that the success of a search based approach relies on how the optimisation is applied and how the fitness of a particular candidate is determined.|
|Appears in Collections:||Dissertations - FacICT - 2017|
Files in This Item:
|6.77 MB||Adobe PDF||View/Open Request a copy|
Items in OAR@UM are protected by copyright, with all rights reserved, unless otherwise indicated.