Using requirements and design information to predict volatility in software development

We hypothesise that data about the requirements and design stages of a software development project can be used to make predictions about the subsequent number of development changes that software components will experience. This would allow managers to concentrate time-consuming efforts (such as traceability and staff training) to a few at-risk, cost-effective areas, and may also allow predictions to be made at an earlier stage than is possible using traditional metrics, such as lines of code. Previous researchers have studied links between change-proneness and metrics such as measures of inheritance, size and code coupling. We extend these studies by including measures of requirements and design activity as well. Firstly we develop structures to model the requirements and design processes, and then propose some new metrics based on these models. The structures are populated using data from a case study project and analysed alongside existing complexity metrics to ascertain whether change-proneness can be predicted. Finally we examine whether combining these metrics with existing metrics improves our ability to make predictions about change-proneness. First results show that our metrics can be linked to the quantity of change experienced by components in a software development project (potentially allowing predictions to take place earlier than before) but that best results are obtained by combining existing complexity metrics such as size, or combining existing metrics with our newer metrics.