Identifying Infeasible GUI Test Cases Using Support Vector Machines and Induced Grammars

Abstract

Model-based GUI software testing is an emerging paradigm for automatically generating test suites. In the context of GUIs, a test case is a sequence of events to be executed which may detect faults in the application. However, a test case may be infeasible if one or more of the events in the event sequence are disabled or made inaccessible by a previously executed event (e.g., a button may be disabled until another GUI widget enables it). These infeasible test cases terminate prematurely and waste resources, so software testers would like to modify the test suite execution to run only feasible test cases. Current techniques focus on repairing the test cases to make them feasible, but this relies on executing all test cases, attempting to repair the test cases, and then repeating this process until a stopping condition has been met. We propose avoiding infeasible test cases altogether by predicting which test cases are infeasible using two supervised machine learning methods: support vector machines (SVMs) and grammar induction. We experiment with three feature extraction techniques and demonstrate the success of the machine learning algorithms for classifying infeasible GUI test cases in several subject applications. We further demonstrate a level of robustness in the algorithms when training and classifying test cases of different lengths.