Measuring Requirement Quality to Predict Testability

Software bugs contribute to the cost of ownership for consumers in a software-driven society and can potentially lead to devastating failures. Software testing, including functional testing and structural testing, remains a common method for uncovering faults and assessing dependability of software systems. To enhance testing effectiveness, the developed artifacts (requirements, code) must be designed to be testable. Prior work has developed many approaches to address the testability of code when applied to structural testing, but to date no work has considered approaches for assessing and predicting testability of requirements to aid functional testing. In this work, we address requirement testability from the perspective of requirement understandability and quality using a machine learning and statistical analysis approach. We first use requirement measures to empirically investigate the relevant relationship between each measure and requirement testability. We then assess relevant requirement measures for predicting requirement testability. We examined two datasets, each consisting of requirement and code artifacts. We found that several measures assist in delineating between the testable and non-testable requirements, and found anecdotal evidence that a learned model of testability can be used to guide evaluation of requirements for other (non-trained) systems.