Investigating NLP-Based Approaches for Predicting Manual Test Case Failure

System-level manual acceptance testing is one of the most expensive testing activities. In manual testing, typically, a human tester is given an instruction to follow on the software. The results as "passed" or "failed" will be recorded by the tester, according to the instructions. Since this is a labourintensive task, any attempt in reducing the amount of this type of expensive testing is essential, in practice. Unfortunately, most of the existing heuristics for reducing test executions (e.g., test selection, prioritization, and reduction) are either based on source code or specification of the software under test, which are typically not being accessed during manual acceptance testing. In this paper, we propose a test case failure prediction approach for manual testing that can be used as a noncode/ specifcation-based heuristic for test selection, prioritization, and reduction. The approach uses basic Information Retrieval (IR) methods on the test case descriptions, written in natural language. The IR-based measure is based on the frequency of terms in the manual test scripts. We show that a simple linear regression model using the extracted natural language/IR-based feature together with a typical history-based feature (previous test execution results) can accurately predict the test cases' failure in new releases. We have conducted an extensive empirical study on manual test suites of 41 releases of Mozilla Firefox over three projects (Mobile, Tablet, Desktop). Our comparison of several proposed approaches for predicting failure shows that a) we can accurately predict the test case failure and b) the NLP-based feature can improve the prediction models.