Improving Automated GUI Testing by Learning to Avoid Infeasible Tests

Most modern end-user software applications are controlled through a graphical user interface (GUI). When it comes to automated test selection, however, GUIs present two major challenges: (1) It is difficult to automatically identify feasible, non-trivial sequences of GUI interactions (test cases), and (2) each attempt at a test case execution can take a long time, eliminating the possibility of rapidly attempting large numbers of alternatives. In this paper we present an iterative approach that infers state-machine models from previous test executions, and increases the utility of tests by learning which sequences to avoid. The approach is evaluated on a selection of Java applications, and the results indicate that our approach is successful at achieving higher code coverage and longer sequences than the state of the art, albeit with a time-overhead caused by the repeated invocation of a Machine Learner.